REVIEW

www.advmat.de

Progress and Opportunities for Machine Learning in

Materials and Processes of Additive Manufacturing
Wei Long Ng,* Guo Liang Goh, Guo Dong Goh, Jyi Sheuan Jason Ten, and Wai Yee Yeong*

In recent years, there has been widespread adoption of machine learning (ML) and make intelligent decisions.[1]
There are four types of ML algorithms
technologies to unravel intricate relationships among diverse parameters in
(Table 1): supervised learning,[2] unsu-
various additive manufacturing (AM) techniques. These ML models excel at pervised learning,[3] semi-supervised
recognizing complex patterns from extensive, well-curated datasets, thereby learning[4] and reinforcement learning.[5]
unveiling latent knowledge crucial for informed decision-making during the Supervised learning learns from a labeled
AM process. The collaborative synergy between ML and AM holds the dataset (input-output pairs) and creates
a mapping between the input data and
potential to revolutionize the design and production of AM-printed parts. This
the corresponding output, allowing the
review delves into the challenges and opportunities emerging at the algorithm to make predictions when pre-
intersection of these two dynamic fields. It provides a comprehensive analysis sented with new data. Although supervised
of the publication landscape for ML-related research in the field of AM, learning requires a lot of human effort and
explores common ML applications in AM research (such as quality control, domain knowledge to label the data and
process optimization, design optimization, microstructure analysis, and define the goal, it can produce accurate
predictions with repeated training itera-
material formulation), and concludes by presenting an outlook that
tions with large relevant dataset, and has
underscores the utilization of advanced ML models, the development of found applications in various industries
emerging sensors, and ML applications in emerging AM-related fields. such as sports[6] and robotics.[7] In con-
Notably, ML has garnered increased attention in AM due to its superior trast, unsupervised learning is trained on
performance across various AM-related applications. It is envisioned that the a dataset without labeled output. It aims
to find patterns or relationships within the
integration of ML into AM processes will significantly enhance 3D printing
large and complex data without human in-
capabilities across diverse AM-related research areas. tervention. Semi-supervised learning is an
ML algorithm that combines the elements
1. Introduction of both supervised and unsupervised learning; it is trained on
a dataset that contains both labeled and unlabeled data to im-
In recent years, the growing interest in machine learning (ML) prove the model performance. It makes efficient use of available
has been driven by a convergence of technological advancements, data and reduces the reliance on costly labeled data to achieve
data availability, community collaborations, and its practical ap- good performance. Lastly, reinforcement learning focuses on
plications in various domains. ML, a subset of artificial intelli- regimented learning processes, whereby the algorithm under-
gence, empowers systems to learn from data, recognize patterns, goes a trial-and-error process based on the provided set of ac-
tions, parameters, and end values to achieve the best possible
W. L. Ng, W. Y. Yeong result.
Singapore Centre for 3D Printing Additive manufacturing (AM), often known as 3D printing,
Nanyang Technological University
50 Nanyang Avenue, Singapore 639798, Singapore
has revolutionized the field of manufacturing by enabling the
E-mail: [email protected]; [email protected] fabrication of customized, complex 3D structures in a layer-by-
G. L. Goh, W. Y. Yeong layer manner. It can be categorized into seven main groups based
School of Mechanical and Aerospace Engineering on ISO/ASTM 52900:2021: 1) binder jetting, 2) directed energy
Nanyang Technological University deposition, 3) material extrusion, 4) material jetting, 5) powder
50 Nanyang Ave, Singapore 639798, Singapore bed fusion, 6) sheet lamination, and 7) vat photopolymerization.
G. D. Goh, J. S. J. Ten When applied to AM, ML opens new avenues for enhancing the
Singapore Institute of Manufacturing Technology (SIMTech)
Agency for Science entire manufacturing process, from material formulation, design
Technology and Research (A*STAR) optimization, and process optimization to quality control. The
5 CleanTech Loop #01-01, Singapore 636732, Singapore synergy between ML and AM has the potential to revolutionize
The ORCID identification number(s) for the author(s) of this article the way AM-printed parts are designed or produced. By harness-
can be found under https://doi.org/10.1002/adma.202310006 ing the vast amount of generated data, ML algorithms can un-
© 2024 The Authors. Advanced Materials published by Wiley-VCH lock deeper insights into AM processes such as optimizing de-
GmbH. This is an open access article under the terms of the Creative signs, predicting material properties, or even improving produc-
Commons Attribution License, which permits use, distribution and
reproduction in any medium, provided the original work is properly cited. tion quality (Figure 1).

DOI: 10.1002/adma.202310006

Adv. Mater. 2024, 2310006 2310006 (1 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Table 1. Different classifications of ML techniques. ergy deposition (59 publications) > vat photopolymerization (20
publications) > binder jetting (8 publications) > sheet lamination
Category Techniques (1 publication). The ML-related AM research has grown substan-
Supervised Learning Linear Regression tially over the last ten years; it has increased significantly from
Logistic Regression
one publication in the year 2013 to 213 publications in the year
2022 (Figure 2b).
Support Vector Machine
As ML is prevalently used in AM processes, further analysis
Decision Trees (e.g., Classification and
was conducted on Web of Science to determine some common
Regression Tree)
ML applications in AM research using the following set of key-
Random Forest
words (“machine learning” + “3D printing/additive manufactur-
Gradient Boosted Trees (e.g., XGBoost,
ing” + “application”). The top five most common ML applica-
LightGBM, CatBoost)
tions in AM research over the last 10 years include 1) Quality
Neural Networks (e.g., DNN, RNN, U-Net,
control (301 publications), 2) Process optimization (222 publica-
RandLA-Net, LSTM)
tions), 3) Design optimization (183 publications), 4) Microstruc-
K-Nearest Neighbours (K-NN)
ture analysis (45 publications), and 5) Material formulation (14
Gaussian Process Modelling
publications) (Figure 2c). These ML applications are applied in
Ensemble learning (e.g., Adaboost, Bayes
optimal classifier, bagging, stacking)
various key AM-related research areas such as aerospace and de-
fense, bioprinting, construction printing, drug printing, electron-
Unsupervised Learning K-means Clustering
ics printing, and marine and offshore and unmanned aerial ve-
Hierarchical Clustering
hicles (Figure 2d). A more in-depth discussion of these five com-
Principal Component Analysis (PCA)
mon ML applications and their specific roles in different key AM-
Independent Component Analysis (ICA) related research areas will be presented in subsequent sections.
Autoencoders (e.g., variational autoencoder)
Gaussian Mixture Models
Semi-supervised Learning Self-training 3. Common ML Applications in AM Research
Multi-view Training
Generative Adversarial Networks (GANs) Over the years, ML has attracted increasing attention in AM
Domain Adversarial Neural Network due to its superior performance in different applications such as
Reinforcement Learning Q-learning quality control, process optimization, design optimization, mi-
Deep Q Networks (DQN)
crostructure analysis, and material formulation. In-depth discus-
sion on the common ML applications will be categorized based
Monte Carlo Methods
on their AM technique to provide a comprehensive overview of
Policy Gradient Methods
how machine learning can be applied to each specific AM tech-
Actor-Critic
nique, thereby highlighting the unique challenges and solutions
that each method presents.

2. Publication Landscape for ML in AM

3.1. Quality Control
An analysis of the publication landscape was conducted to de-
termine the influence of ML on different 3D printing techniques Quality control plays a crucial role in enhancing the efficiency
using the following set of keywords on Web of Science (“machine and reliability of additive manufacturing processes. Many studies
learning” + “3D printing/additive manufacturing” + “printing have explored the application of ML algorithms and sensor data
technique”). As there are many variants of 3D printing tech- analysis to achieve real-time process monitoring and quality as-
nique under each ASTM classification, numerous keywords for surance. Signals from in situ sensors are used to train ML models
each printing technique were used: 1) binder jetting—binder jet- to monitor the stability of the process and detect defects within
ting, multi-jet fusion; 2) directed energy deposition—directed successful builds. Different ML techniques (supervised, unsuper-
energy deposition, wire arc additive manufacturing; 3) material vised, semi-supervised, and reinforcement learning) have been
extrusion—extrusion, fused deposition modeling, fused filament used for quality control in AM; the choice of ML approach for
fabrication, direct ink writing; 4) material jetting—jetting, inkjet, quality control is dependent on the nature of the data and the
microvalve; 5) powder bed fusion—powder bed fusion, selective objectives of the quality control system. Supervised learning is
laser sintering, selective laser melting, electron beam melting, most used ML technique for quality control in AM and most of
laser powder bed fusion; 6) sheet lamination—sheet lamination; the studies demonstrated high prediction accuracies > 90%. Al-
and 7) vat photopolymerization—vat photopolymerization, stere- though it is well-suited for quality control with labeled historical
olithography, digital light processing (DLP), and continuous liq- data from both good and defective products/processes, it may not
uid interface production. As shown in Figure 2a, a total of 528 be useful for the detection of novel defects or anomalies not seen
ML-related AM publications were published over the last 10 years in the training data. Figure 3 provides a summary of the tech-
and the adoption of ML for each 3D printing technique varies— niques involved and applications of ML in quality control and
powder bed fusion (225 publications) > material extrusion (135 more discussion on the different types of ML used in quality con-
publications) > material jetting (80 publications) > directed en- trol will be provided in the subsequent sections.

Adv. Mater. 2024, 2310006 2310006 (2 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 1. An overview of the integration of ML into AM processes for diverse optimizations and applications. The left section classifies ML into four
types—supervised, unsupervised, semi-supervised, and reinforcement learning—while introducing the emerging transformer model for ML applica-
tions. On the right, various AM techniques are detailed. The central part of the figure illustrates the potential benefits that ML can offer to AM and the
bottom part showcases practical applications across a wide range of industries, from aerospace and defense to electronics, and food, underscoring the
extensive impact of integrating ML with advanced manufacturing methods. Reproduced with permission.[53,76,115,176] Copyright 22 Mar 2024, Elsevier.

Adv. Mater. 2024, 2310006 2310006 (3 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 2. a) Number of ML-related additive manufacturing publications over the past 10 years. b) Detailed annual breakdown of the number of publi-
cations for each printing technique from years 2013 to 2022. c) Detailed annual breakdown of the number of publications for different ML applications
in AM from years 2013 to 2022. d) Detailed annual breakdown of the number of publications on ML applications in various key AM research areas from
years 2013 to 2022.

3.1.1. Powder Bed Fusion To monitor the process stability, a study used data from the
supplied EOS M290 powder bed images to predict anomalies dur-
It is common to observe high porosity, balling, incomplete fu- ing the powder spreading process (Figure 4).[9] The powder-based
sion, and spattering during the AM process. These defects arise materials included Ti6Al4V, AlSi10g, IN718, SS316L, SS17-4, and
from factors such as process instability and poor material interac- bronze. The images were first filtered using 37 different 2D im-
tions. Such defects are detrimental to the quality of printed parts, age processing filters, and the filter responses were stored in vec-
leading to compromised mechanical strength, surface rough- tors for each pixel. The response vectors were then grouped into
ness, inaccurate geometries, and potential delamination. It is im- 100 groups using a standard k-means unsupervised clustering
portant to address these defects to ensure the reliability and per- algorithm. The mean response vector for each group was then
formance of AM-fabricated components across industries.[8] stored as visual words in a dictionary. Then, the pixel at each
training image patch was matched to the closest visual word,
and the histogram for the occurrence frequency of each word
in the patch was calculated and termed “fingerprints”. The ra-
tionale was that training images with similar powder-spreading
anomalies would result in similar “fingerprints”. During the
method execution, the powder bed image was divided into dif-
ferent patches and the “fingerprint” from each patch was then
calculated. The quality of a patch was determined by matching
its “fingerprint” to a database of 2402 “fingerprints”. These “fin-
gerprints” were manually labeled under six conditions: anomaly-
free: 1040 “fingerprints”, recoater hopping: 264 “fingerprints”, re-
coater streaking: 228 “fingerprints”, debris: 187 “fingerprints”,
super-elevation: 314 “fingerprints”, part failure: 264 “finger-
prints”, incomplete spreading: 105 “fingerprints”. The top three
matches from this database were then used to assess the patch’s
quality. The algorithm managed to classify the powder spreading
Figure 3. Graphical overview summarizing the diverse applications of ML based on the six conditions with precision ranging from 65.0%
in quality control across various AM processes. to 98.9%

Adv. Mater. 2024, 2310006 2310006 (4 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 4. Flow chart of ML process to detect anomaly in laser powder bed fusion. Reproduced with permission.[9] Copyright 2018, Elsevier.

For defect detection in laser powder bed fusion (LPBF) builds, In-process monitoring using infrared cameras followed by ML
various sensor technologies were used in conjunction with ML for data analysis was performed using the original equipment
techniques. These technologies included visible light cameras, manufacturer and customized hardware.[12] The EOSTATE Expo-
infrared cameras, high-speed cameras, photodiodes, and acous- sure OT captured a long exposure image at ≈900 nm of the laser
tic sensors. There were also efforts that combined multiple sen- scanning over one whole layer. Both groups changed the process
sor technologies for ML training. Using a high-resolution 36.3- parameters to create the training dataset. The unsupervised K-
megapixel digital single-lens reflex (DSLR) camera, images of means clustering was used to enlarge the manually labeled train-
each layer were taken before and after LPBF laser scanning to ing dataset followed by k-nearest neighbors (K-NN) supervised
predict the locations of voids.[10] Multiple images were taken un- learning to identify anomalies (drifts) in the images.[12a] These
der different lighting conditions for each build layer and were anomalies were then shown to have a correlation with high poros-
combined using an ensemble classification. It was possible to com- ity occurrence in X-ray CT scans of the actual samples. Random
bine data from multiple sensors through the ensemble technique forest-bagged tree ensemble labeled with X-ray CT data are used
instead of only using data from a single sensor under different (Figure 5) and the use of multiple consecutive layers improved
conditions. The ground truths were obtained from X-ray com- the prediction accuracy. The interpretability of the random forest
puted tomography (CT) convolved with a Gaussian filter and la- (RF) model showed that the lack of fusion defects prediction was
beled using support vector machine (SVM) binary classifiers be- dependent on the adjacent layers while keyhole defects prediction
fore being manually checked by a certified non-destruction in- was heavily dependent on the 10th subsequent layer. The model
spection inspector. The ensemble method improved the accuracy could determine the average density of a small area measuring
of prediction from 65% to 85% for single sensor image input to 1 mm × 1 mm.
85% for multiple images. An infrared thermographic camera was integrated to an SLM
A four-phase (sliding, convolutional neural networks 280 LPBF system to capture short videos of delamination, splat-
(CNN), smoothing, and compensation) modeling approach ters, and defect-less processes that were then converted to image
was developed for online surface measurement in additive frames to train a CNN.[13] The training dataset was augmented
manufacturing.[11] This approach utilized a window-based data by rotating the image, flipping the images, and random image
reformulation technique and CNN to predict 3D surface data noise and blur. Delamination and splatter defects were detected
directly from 2D images without the need for time-consuming at an average accuracy of 96.8%.
triangulation computations. The method proved to be highly High-speed cameras with frame rates above 1000 Hz in the
accurate, with an average relative prediction error mostly lower visible and infrared range were also used as inputs for ML mod-
than 10%. Its computational efficiency and ability to acquire els. A short wavelength infrared high-speed camera was used to
data layer-wise in real-time made it suitable for online quality capture the thermal history of the part and train a customized
monitoring and control in additive manufacturing processes. CNN for the identification of defect locations within the part.[14]

Adv. Mater. 2024, 2310006 2310006 (5 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 5. Workflow of training the ML model using labeled X-ray CT data. Initially, a compilation of through-process data encompassing Computer-Aided
Design (CAD), processing parameters, real-time online monitoring data captured via Optical Tomography (OT) images, and subsequent post-processing
characterization data derived from X-ray Computed Tomography (CT) was performed. Subsequently, an ML model was developed to glean valuable
insights into the mechanisms underlying defect generation. The adeptly trained ML model is proficient in accurately forecasting porosity occurrences
within individual layers, leveraging the composite data from multiple layers of OT information. Reproduced with permission.[12b] 2022, Elsevier.

Melt-pool and time-dependent attributes were extracted from the morphology, spatter characteristics, and melt pool temperature
thermal images for groups of pixels and a threshold was set for features).
a binary outcome of pore and non-pore groups. The data was la- One of the challenges for training ML models with high-speed
beled in comparison to X-ray CT data with details from their pre- camera data was the difficulty in volumetrically matching the
vious work[15] and a 1D CNN model was trained using Bayesian ground truth data typically obtained by X-ray CT. Synchrotron X-
Optimization. It was found that a group of pixels representing ray imaging was used to obtain real-time defect formation data
a volume of 700 × 700 × 50 μm produced the best prediction for comparison with the in situ visible and NIR imaging above
accuracy for keyhole porosities above 0.1% in volume and de- 50 kHz instead of measuring the defects after the process.[18]
creasing the volume size reduced the prediction accuracy. A high- The frequency response of the intensity of the high-speed cam-
speed visible light camera at 6,400 Hz was used to capture melt era images was grouped into wavelets shorter than 1 ms for train-
pool images.[16] The feature extraction, classification, and sub- ing a deep neural network model to identify pore and non-pore
sequent unsupervised ML model training were similar to an- events. Simulation models were then used to provide further in-
other work for build failure detection and modifications were sights into the pore formation mechanism. The developed simu-
made to achieve a scale-invariant representation of the melt pool lation model was then used to generate data to train a deep learn-
morphology.[9] The “fingerprints” that were identified for five out- ing model for a high-speed camera integrated into a commercial
comes include desirable, under-melting, keyhole porosity, severe LPBF system SLM 280. Accuracies up to 87% were obtained for
keyhole porosity, and balling. Two infrared (IR) high-speed cam- identifying the pores.
eras at 700 and 950 nm wavelengths respectively were used to Due to the limited penetration depth of visible and IR waves
take images of melt pools at 100 kHz (Figure 6).[17] Multiple in metals, the signals obtained from electromagnetic emissions
feature types from the raw sensor data and calculated thermal of these wavelengths typically only capture signals from the sur-
field were extracted to train various ML models ranging from K- face of the metal during the process. Besides using synchrotron
NN, SVM, to CNN. The algorithms were used to 1) detect out-of- X-ray to penetrate the metal, others have captured acoustic emis-
focus laser and 2) porosity level and were able to achieve a true sions to potentially detect process signals originating from below
positive rate of 90%. The computationally light ML models pro- the metal surface. A microphone was secured 25–30 cm on top
duced subpar results when trained on single feature types but of the build plate to sample acoustic emissions at 100 kHz.[19]
generated on-par results with the deep learning models when Features from the signal were extracted based on three primary
trained on inputs from the multiple feature types (melt pool groups: time-series statistics, frequency domain characteristics,

Adv. Mater. 2024, 2310006 2310006 (6 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 6. Diagram illustrating the diverse physics-derived attributes extracted from various sensing methods, which are then used to train the CNN
architecture. A visual depiction of the efficacy of various models in classifying a) the size of the laser spot and b) the kind of porosity, measured using
the F1-score. RRC stands for ridge regression classifier, NLR denotes nonlinear logistic regression classifier, SVM represents support vector machine
classifier, and CNN signifies convolutional neural network classifier. Adapted with permission.[17] 2022, Elsevier.

and oscillatory modes via an ensemble empirical mode decom- means clustering achieved a 90% prediction rate when paired
position technique. An SVM was then trained based on X-ray with a deep learning classifier, cracks were detected using prin-
radiograph ground truths for signal windows of 1–15 ms. A cipal component analysis and lastly the general deep learning
window of 7.5 ms showed the best accuracy of 97% in pre- model (trained on raw H13 signals) demonstrated good adapt-
dicting keyhole pore and keyhole-free scan lengths that corre- ability for prediction when tested on SS316L data. A 1D CNN
sponded to the window. A sensor measuring a wideband of model accurately detected the spattering event in the LPBF pro-
100–900 kHz at the center bottom of a circular build plate was cess up to 85% (Figure 7a).[21] Conversely, acoustic emission from
used to detect acoustic emissions from coupons separated ra- the powder bed fusion process was used to predict possible de-
dially from the sensor.[20] The noise from the acoustic signal fect formation within the printed parts (Figure 7b).[22] The devel-
was first removed and three methods were explored: k-means, oped model was able to predict different types of defects (lack
principal component analysis, and a general deep learning. K- of fusion pores, conduction mode, and keyhole pores) within

Adv. Mater. 2024, 2310006 2310006 (7 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 7. a) Schematic showing the ID-CNN model that is used for detecting spattering event using acoustic signal. Reproduced with permission.[21]
2021, MDPI. b) Schematic showing the flowchart of model that is used for detecting various types of defects using acoustic signal. Reproduced with
permission.[22] 2022, Taylor & Francis.

different materials (316L stainless steel, bronze (CuSn8), and In- demonstrated high accuracy, further validation is necessary for
conel 718) with an accuracy of ≈93%. It was challenging to col- complex geometries and scanning paths, other types of de-
lect the acoustic emissions as the signal passed through the pre- fects, and optimization of hardware and data collection pipeline,
viously melted layers and the build plate to reach the sensor. and the inclusion of physics-based inference from the trained
Another study collected the acoustic signals via an optoacous- models.
tic fiber Bragg grating with sampling rates up to 10 MHz.[23] The previous studies were either based on derived or raw data
Reinforcement learning was used to recognize acoustic emis- taken from one sensor type. There were also efforts to use ML
sions from three classes of material: poor quality, medium qual- for multiple sensor types to achieve improved defect detection
ity, and high quality, and achieved detection accuracies above rates. Sensor data from the optical layer images, process multi-
74%. A monitoring strategy for LPBF process was developed us- spectral emission, and the vector scan path of the laser were com-
ing a hybrid deep learning (DL) model that combined CNN and bined (Figure 9).[25] The combined sensor data input was trained
long–short-term memory (LSTM) (Figure 8).[24] The proposed against X-ray CT ground truths in a CNN model to classify vol-
model achieved high prediction accuracy ranging from 95.9% to umes of 940 μm × 940 μm × 660 μm into flaw and nominal
100% in classifying lack of fusion, conduction mode, and key- build regions. The training accuracy was 97.3% when trained
hole across various time scales, based on data from a heteroge- solely on multi-spectral emissions but decreased to 88.7% for
neous time-synced sensing system. The study emphasized the the test dataset. Similarly, the training accuracy was 97.0% when
importance of back reflection and structure-borne acoustic emis- trained on all data modalities but decreased to 91.9% for the test
sion sensors in the decision-making process. Although the model dataset.

Adv. Mater. 2024, 2310006 2310006 (8 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 8. Overview of the variable time scale monitoring of LPBF using a hybrid DL model. The proposed DL model can operate over variable time
scales for LPBF monitoring. A hybrid DL architecture combining CNN and LSTM was introduced. Heterogeneous signals, including optical and acoustic
emissions from the process zone, were used to train the DL model. The DL model demonstrated accurate classification of lack of fusion, conduction,
and keyhole regimes within time scales ranging from 0.5 to 4 ms. Adapted with permission.[24] 2022, Elsevier.

3.1.2. Material Extrusion a minimum accuracy of 95% with both small and large datasets.
This ML algorithm provided faster and cheaper quality assurance
Material extrusion, a prevalent additive manufacturing tech- compared to traditional optical 3D scan methods.
nique, is prone to specific defects that impact the quality of Nozzle clogging is also a common issue when dealing with
printed parts. Common defects include layer misalignment fiber-reinforced polymers. Nozzle clogging is usually undetected
caused by inaccurate deposition, voids stemming from incom- by the printer and would cause an eventual print failure. A multi-
plete material fusion, inconsistent extrusion leading to irregu- head encoder-decoder temporal convolutional network (MH-ED-
lar wall thickness, and delamination due to weak adhesion be- TCN) algorithm utilized time-series data from collaborative sen-
tween layers. These defects are dependent on factors like im- sors to detect nozzle clogging.[27] This algorithm outperformed
proper temperature control, inadequate material flow, and incor- other ML approaches including SVM, LSTM, LSTM autoencoder,
rect print settings. These undesirable defects compromise the and a simple CNN to achieve a remarkable 97.2% accuracy in
structural integrity, mechanical strength, and dimensional accu- identifying nozzle clogging. Further improvements to the algo-
racy of printed parts, making defect mitigation essential for pro- rithm were also recommended and the addition of appropriate
ducing reliable and functional components. sensors addressed the printing malfunctions caused by the vis-
To improve print quality and consistency of fused filament fab- coelastic behavior of polymer materials. Although prediction of
rication (FFF) processes, an innovative approach to quality assur- nozzle clogging can be performed using ML approaches, it is
ance in additive manufacturing processes was implemented by challenging to predict the quality of extruded materials from the
leveraging environmental data and ML.[26] Various environmen- nozzle. To solve that, image-based anomaly detection techniques
tal parameters (temperature, humidity, air pressure, and gas par- were developed to realize real-time monitoring and correction.
ticles) were recorded and analyzed during fused deposition mod- Image classification model[28] and object detection models[29]
eling (FDM) processes and different ML algorithms (multilayer were used to predict under-extrusion and over-extrusion phe-
perceptron (MLP), 1D CNN, RNN, LSTM, Inception Time, Xcep- nomena during the FDM process and accuracies of 98.0% and
tionTime, and eXplainable CNN for multivariate time series clas- 89.8% were achieved respectively. The system outperformed hu-
sification (XCM)) were employed for classification. The Xception- man response times in detecting and correcting defects. The
Time architecture was found to be the most effective, achieving framework proposed can be extended to other 3D printing

Adv. Mater. 2024, 2310006 2310006 (9 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 9. A Convolutional Neural Network (CNN) design for defect identification uses 3D image slices from M different sensor types. This network is
structured with two convolutional phases followed by a dense layer comprising one hidden layer. The model’s output predictions are produced through a
softmax classification layer. Bar chart showing sensor fusion performing better than the individual layer-wise optical layer images, process multi-spectral
emission, and the vector scan path of the laser. Reproduced with permission.[25] 2022, Elsevier.

technologies for fabricating high-performance materials in chal- Defect detection is critical for large-scale printing due to the
lenging environments without human intervention. high cost involved, especially so in building and construction.
A study was performed to diagnose faults and identify causes, A study has demonstrated automated layer defect detection in
particularly regarding the drift of process parameters. A deep ad- construction 3D printing using deep CNN.[32] The system com-
versarial learning system that utilized captured upper-layer im- prised a deep CNN model that took images as input and distin-
ages during the manufacturing process was proposed.[30] It em- guished concrete layers from surrounding objects via semantic
ployed a conditional generative adversarial network (CGAN) to pixel-wise segmentation. Data augmentation techniques gener-
address data imbalance and a domain adversarial neural net- ated 1 million images for training, tuning, and testing the CNN
work (DANN) to handle domain-shifting problems caused by model. Furthermore, a defect detection module was developed to
drifting process parameters. The experimental validation demon- detect deformations in the printed concrete layers using the im-
strated the effectiveness and accuracy (91.01%) of the proposed ages output by the CNN model. The evaluation results showed a
method. Although many approaches have been developed to pro- high level of accuracy and F1 score (>90%) in differentiating con-
vide automated monitoring, current automated methods can- crete and non-concrete pixels, while the defect detection module
not be universally applied to various components, materials, achieved a total accuracy of 97.5% and a miss rate of less than
and printing systems. A study focused on generalizing the 3D 6% for printed layers with and without defects. A similar study
printing defects and correcting errors in material extrusion ad- with fewer images resulted in poorer performance of 80% mean
ditive manufacturing (Figure 10).[31] A multi-head neural net- average precision.[33] These studies demonstrated the potential
work trained on a large and diverse dataset (≈1.2 million im- of computer vision and deep learning techniques for automated
ages) was developed to identify deviations from optimal print- inspection and quality monitoring in construction 3D printing.
ing parameters. The system allowed for real-time error detec- In construction printing, the bigger size (cm-scale) of the ex-
tion and rapid correction across different printing scenarios. The trudate allowed the utilization of 3D scanners for detecting defor-
trained network achieved an overall accuracy of 84.3% in clas- mations in printed structures.[34] This was unlike smaller-scale
sifying the flow rate, lateral speed, Z offset and hot end tem- polymer printing (in mm scale) where the precision of 3D cam-
perature and demonstrated the effectiveness of gradient-based eras imposed limitations on this capability. A study evaluated the
visual explanations for understanding network decisions. The performance of a monocular camera, LiDAR, and LiDAR-camera
methodology offered a cost-effective and scalable solution that in terms of point cloud density and 3D map reconstruction for
can be easily integrated into existing printers and workflows, defect detection. The results showed that the RGB-L camera out-
leading to improved quality and reliability of end-use products performed the other sensors in all scenarios, with an error be-
(Figure 11). low 4% when using K-means clustering at a distance of 0.5m.

Adv. Mater. 2024, 2310006 2310006 (10 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 10. a) The feedback pipeline consists of six key steps that facilitate the online updating of parameters based on image data obtained during
the extrusion process. b) The provided table presents the values for 𝜃 mode (mode threshold), L (sequence length), Imin (interpolation minimum), A+
(maximum increase), and A− (maximum decrease) for each printing parameter, along with the corresponding possible levels of update amounts. c) A
simple example is presented to illustrate the geometric structure of a single layer and the subdivision of the toolpath into smaller segments of equal
length. This subdivision, using 1 mm segments, enables swift correction and reduces the response time in the feedback process. Reproduced with
permission.[31] 2022, Nature Portfolio.

However, the execution time of the algorithm was currently too rial) within each printed layer is compounded with increasing lay-
high for real-time applications. It was suggested that future work ers and it would lead to poor dimensional accuracies for large 3D-
can focus on validating the system by analyzing concrete printed bioprinted tissue constructs, which typically require a long print-
areas, incorporating color information to better identify points ing time and involve high material cost. The CNN-based classi-
belonging to each printed layer, and developing the more ad- fiers are typically utilized for defection detection in most man-
vanced algorithm in C/C++ to reduce computational cost and en- ufacturing processes; they can be implemented to monitor and
able real-time applications. improve the printing outcome in 3D bioprinting processes using
A novel methodology for real-time quality assurance in 3D- computer vision. The captured images can be labeled as “under-
printed electronics using U-Net was presented.[35] An FFF printer extrusion”, “good-quality” and “over-extrusion” images for train-
equipped with an extruder was used for conductive paste dispens- ing. A DL model can be used to optimize the printing parame-
ing, pick-and-place unit, and dual cameras. The cameras cap- ters iteratively and adaptively using a real-time in situ monitoring
tured images during the printing process and a trained neural and correction system. An ad hoc optimized CNN and a mathe-
network was used to distinguish the conductive wires from the matical model were used to perform in-process and parameter
plastic substrate. The method was used to identify common print- optimization of the extrusion-based bioprinting process.[36] The
ing flaws such as connection breaks, shorts, and inaccuracies in dataset was constructed by capturing videos of multi-layered scaf-
wire width, comparing the actual output with the intended G- folds fabricated using the extrusion bioprinting process; the in-
code instructions with an overall accuracy of 96.6%. The results puts include type of extrusion system (pneumatic or mechanical),
facilitated high-resolution documentation and provided data to type of material, layer thickness, and infill density while the out-
improve the printing process. This innovation enabled the de- put is based on extrusion multiplier (which represents the ratio
tection of errors and can be potentially used for automated flaw of printing resolution to nozzle diameter). The printing quality
rectification, paving the way for more reliable and autonomous can be optimized by tuning the printing parameters through a
3D-printed electronics production. series of consecutive prints in a feedback loop manner using the
In situ monitoring is commonly implemented during the bio- CNN model. The results showed an accuracy of 94.3% for overall
printing process to improve the dimensional accuracies of 3D- printing; acceptable printing has a precision of 87.2% and recall
bioprinted tissue constructs. The error (missing or excess mate- of 96.5%, over-extrusion has a precision of 98.3% and recall of

Adv. Mater. 2024, 2310006 2310006 (11 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 11. a) The multi-head neural network allows for quick correction of errors caused by manual intervention in a single parameter. It has been
trained on a particular printer and PLA feedstock. The correction procedure is used on a hidden 0.4 mm nozzle that was not part of the training set
of data. b) The control pipeline shows that multiple incorrect parameters for thermoplastic polymers that were not seen during the training phase can
be simultaneously optimized online. This demonstrates the system’s adaptability to a variety of feedstocks with various material characteristics, colors,
and initial conditions. c) The system uses self-learned relationships between parameters to make corrective predictions, much like human operators
do. By decreasing the Z offset value and/or increasing the material flow rate, for example, a high Z offset can be corrected. d) The system successfully
fixes numerous wrong printing parameters that were added during a print job. The only difference between the two identical rooks printed under the
same circumstances was how the correction process was used. e) Prints started with the wrong parameter combinations are successfully handled by the
system. The same conditions were used to print a set of six spanners, demonstrating the system’s capacity to correct mistakes and produce the desired
results. Reproduced with permission.[31] 2022, Nature Portfolio.

94.5% and lastly under-extrusion has a precision of 97.6% and and shape. An MLP-based non-linear autoregressive model was
recall of 92.2%. trained to predict droplet size and velocity with a statistical fi-
These research efforts collectively demonstrate the growing delity exceeding 90%, outperforming traditional statistical mod-
potential of ML and computer vision in automating quality in- els. Furthermore, a supervised ML model was trained to classify
spection and monitoring in a material extrusion process. By ad- droplet shapes using spectral frequencies from the MW sensor
dressing the limitations of manual inspection methods, these data, achieving an F1-score of over 95%. This approach presented
advanced solutions offer enhanced accuracy, efficiency, and the a practical and computationally efficient solution for quality con-
potential for real-time applications. Further developments in trol in liquid metal jetting AM. It was even suggested that future
dataset size, sensor technology, and algorithm optimization hold research should aim to develop and use ML models for the pre-
promise for broader adoption and improved quality control in the diction of defect formation and build failures and contribute to
field of AM. improved part quality and higher manufacturing efficiency.
A novel in situ monitoring method employing vision-based
techniques was introduced to observe droplet formation in inkjet
3.1.3. Material Jetting printing.[38] A drop watcher camera was implemented to capture
video sequences of droplet properties which include size, veloc-
A technique for in-process monitoring of droplet properties dur- ity, aspect ratio, and the existence of satellite droplets under vari-
ing liquid metal jetting AM was developed using an in-process ous voltage and frequency combinations. The influence of these
millimeter wave (MW) sensor and ML.[37] The MW sensor pro- parameters on distinct droplet modes (namely normal, satellite,
vided a real-time monitoring solution that circumvented the com- and no-droplet) was analyzed through computer vision, and a
puting requirements of high-speed image sensors by produc- backpropagation neural network (BPNN) was constructed to cat-
ing efficient time series data to anticipate droplet size, velocity, egorize the droplet modes based on these properties with a high

Adv. Mater. 2024, 2310006 2310006 (12 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 12. Illustration of the process for detecting irregular powder feeding. a) Deposition head featuring four side nozzles and a central camera for
monitoring the melt pool region, b) Training and validation of the model using image datasets representing both regular and irregular conditions, and
c) Real-time application of the pre-trained model for immediate inference. Reproduced with permission.[43] 2022, Elsevier.

degree of classification accuracy at 90%. This method offered a ing the importance of meticulous parameter control and real-
sturdy framework for real-time quality inspection during inkjet time monitoring in DED processes for achieving high-quality
printing, which can potentially facilitate process enhancement components.
and predictive analysis. The work laid the foundation for the fu- An infrared camera coaxial to the laser beam was used to
ture development of a digital twin model for inkjet printing and train a deep CNN to identify process stability at four categories:
other related electronic printing technologies. normal laser power, low laser power, low scanning speed, and
Another study proposed the adoption of predictive models high scanning speed with accuracies above 80%.[41] Another
and nonlinear autoregressive neural networks with external in- study analyzed process parameters to predict the melt pool tem-
put (NARX) for quality assurance and process control in AM peratures via extreme gradient boosting ensemble learning and
for electronics. The challenges of using 3D printing in elec- LSTM neural networks.[42] The melt pool temperatures were
tronics manufacturing were highlighted and modeling tech- trained on infrared camera measurements and the coefficient
niques such as finite element analysis (FEA) and data-driven of determination between the prediction and actual measured
ML can be applied for predicting product performance, qual- temperatures was higher than 51% in all cases tested. Another
ity, and reliability.[39] A novel model-based approach for inkjet study used coaxial camera images to determine abnormal pow-
printing process is demonstrated using state-space models de- der supply in DED due to issues such as restricted powder flow
rived from measured process data.[40] This approach helped to (Figure 12).[43] A few ML models were trained and the accura-
anticipate process trends and associated product quality charac- cies were above 55.1%, 69.8%, 70.6%, and 95.9% for K-NN, de-
teristics over large prediction horizons, even in the case of mod- cision tress (DT), RF, and CNN, respectively. Furthermore, an
erately non-linear dynamics of the 3D printing process. Both in situ monitoring system was developed for DED process.[44]
studies emphasized the importance of proactive, model-based The monitoring system consisted of a hyperspectral camera for
assessment over conventional post-manufacture techniques to melt pool width and temperature control, a coaxial camera for
mitigate common reliability and quality risks associated with melt pool data, and a laser scanning system for material height
AM. These advancements have significant potential in enhanc- measurement. Independent tuning of the process parameters
ing the acceptance of 3D printing technology in the electron- was challenging due to the highly interconnected process param-
ics industry, while ensuring improved and more robust process eters that significantly influenced the deposited geometry and
performance. material properties. Hence, ML-based optimization is helpful in
finding the optimal controller output for the in situ process mon-
itoring system.
3.1.4. Directed Energy Deposition (DED) Beyond process stability, neural network-based ML such as
RandLA-Net with in situ sensors were used to detect defects
The occurrence of defects is common in the DED process; these within the DED built parts. Geometrical defects were classified
defects include balling, lack of fusion, porosity, warping, and based on a laser line scanner trained on a DL model to clas-
waviness. The defects are caused by improper laser parame- sify surfaces into normal, convex, and concave at accuracies of
ters, material interactions, and heat accumulation. These is- 91.3%.[45] A similar study used data from a laser profiler to first
sues compromise the structural integrity, surface finish, and cluster the points via an unsupervised ML model followed by su-
dimensional accuracy of the manufactured parts, highlight- pervised learning to classify the clusters into no defect, bulging,

Adv. Mater. 2024, 2310006 2310006 (13 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

dents, and wavy surfaces.[46] For the supervised learning, a few

ML models were explored, and K-NN achieved the highest ac-
curacy of 93.2%. Another work combined optical emission spec-
trometer and CCD camera images through Kronecker product of
graphs as inputs into an SVM model.[47] The model was trained
against X-ray CT data to classify the layers into three categories:
low, medium, and high severity. The combination of input from
the two sensors via a Kronecker product improved the statisti-
cal fidelity score from 35% to 75%. A microphone with a sam-
pling rate of 44.1 kHz was used to gather acoustic emissions
from the DED.[48] A deep learning model that consisted of a
fully connected regression deep neural network (F-DNN) and a
redundant convolutional encoder-decoder network (R-CED) was
trained to remove the signal noise from the machine motion, in-
ert gas motion, and powder supply based on the ground truth Figure 13. Graphical overview of ML applications in optimizing processes
signal that was equalized, filtered and processed using audio- across diverse AM processes, along with various objectives related to pro-
cess optimization.
based algorithms.[48a] The acoustic signal was first denoised by
audio-based signal processing algorithms, followed by extraction
of time and frequency features to construct a sequence of Mel- data (labeled or unlabelled), and the level of complexity involved
Frequency Cepstral Coefficients (MFCC).[48b] These coefficients and more discussion will be provided in the subsequent sections.
were used as inputs to train a CNN model and compared against
simpler ML models such as RF, SVM, gradient boosting, and K-
NN with a subset of the audio signal features used as inputs. The 3.2.1. Powder Bed Fusion
MFCC CNN model produced the highest accuracies above 89%
at predicting defect free, cracks, and keyhole pores in the built The quality of LPBF-fabricated parts is heavily influenced by
part (Table 2). process parameters, but existing methods for determining the
parameter window are time-consuming and subjective. A su-
pervised ML method was implemented to optimize the LPBF
3.2. Process Optimization process in additive manufacturing (Figure 14).[53] An ML ap-
proach was proposed to detect and track defects and predict ma-
Process optimization plays a crucial role in maximizing the effi- terial printability in LPBF. It classified printed tracks into five
ciency and reliability of AM processes. With the complexity and groups based on surface characteristics and developed a data-
intricacy involved in AM, fine-tuning the printing parameters, driven model using BPNN. The model utilized the classification
and optimizing the process parameters are essential for achiev- results as target output and four quantitative indicators calcu-
ing consistent and desirable results. One of the major challenges lated from surface morphology as input variables. The proposed
in process optimization for 3D metal printing is the intricate in- method significantly improved the efficiency of parameter win-
terplay of numerous variables, including laser power, scanning dow search, enabling defect-free printing and excellent part per-
speed, layer thickness, and powder characteristics. The optimal formance. The integration of a 3D microscope for in situ mea-
combination of these parameters is dependent on the specific surement further enhanced its applicability in unmanned facto-
metal alloy, part geometry, and desired mechanical properties. ries. Overall, the work highlighted the importance of ML in op-
This is where ML is important in process optimization for metal timizing the LPBF process, offering an intelligent solution for
printing. ML algorithms can analyze large amounts of data from parameter determination and paving the way for more efficient
previous printing runs, identifying patterns and relationships and automated manufacturing processes.
between process parameters and part quality. By learning from The primary object of another study was to determine the opti-
these patterns, ML models can predict the optimal process pa- mal laser tool path by minimizing the average thermal gradient.
rameters for a given set of conditions, thereby reducing the need The study showcased the capability of accurately predicting opti-
for trial and error, and minimizing material waste. In general, mal laser paths using a DL model, which was trained on 33000
ML can be implemented in metal printing to predict printability physics-based simulation results that contain “good” labels (low-
of a material under specific set of process parameters, to opti- temperature gradient) and “bad” labels (high-temperature gra-
mize the toolpath to decrease the residual stress, and to predict dient) in a 1:1 ratio, despite limitations in training data and bi-
the thermal gradient for process optimization purposes. nary information.[54] Notably, the DL simulation, implemented
Supervised learning is commonly used for process optimiza- with a CNN, significantly outperformed brute force simulations
tion in AM; it is suitable for process optimization when histori- in terms of speed. This work underscored DL’s potential in tool
cal data with well-defined input-output pairs are available, but it path optimization within AM, highlighting its ability to compre-
may not adapt well to dynamic processes. Figure 13 provides an hend tool path patterns and reconstruct comprehensive path per-
overview of ML applications in optimizing processes across di- formances. Furthermore, the research illustrated the feasibility of
verse AM processes, along with various objectives related to pro- applying a physics-based DL approach to other AM techniques,
cess optimization. The choice of ML approach is often dependent providing lower simulation costs while maintaining accuracy.
on the specific process optimization problem, the availability of Thus, a trade-off between computational expenses and accuracy

Adv. Mater. 2024, 2310006 2310006 (14 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 Table 2. ML for quality control in AM.

Research target Fabrication process ML technique Sample size Inputs Outputs Performance Refs.

Process stability: anomalies PBF Unsupervised: 2402 image patches Powder bed images Powder spreading Able to classify the [9]
in powder spreading K-means unsupervised clustering quality: anomaly-free, powder spreading

Adv. Mater. 2024, 2310006

followed by labelling based on recoater hopping, with precision
expert knowledge of the powder recoater streaking, ranging from 65 to
spreading defects debris, 98.9%.
super-elevation, part
failure, or incomplete
www.advancedsciencenews.com

spreading
Defect prediction: location PBF Supervised: 840 samples of Build layer images taken Binary quality of a Prediction accuracy: [10]
of voids Ensemble classification DSLR voxels with DSLR camera at group of voxels as a 85%
eight different flaw or nominal build
lighting conditions
Layer-wise Surface PBF Supervised: 70 000 samples 300 × 300-pixel images Surface morphology Average relative [11]
morphology CNN prediction error lower
measurement than 10%

2310006 (15 of 56)

Defect prediction through PBF Unsupervised: 200 Long exposure IR Labelling of in situ IR Accuracy close to 100% [12a]
labelling of anomalous K-means clustering to enlarge the camera images camera images as
in situ data manually labelled training (OEM EOSTATE either normal or
dataset k-nearest neighbours Exposure OT) anomalous
supervised learning to identify
anomalies
Defect prediction through PBF Supervised: 100 000 to 1 000 000 Long exposure IR Average porosity of a Accuracy greater than [12b]
average porosity in a Random forest bagged tree camera images local volume 90%
local volume ensemble. (OEM EOSTATE
Exposure OT)
Defect prediction of PBF Supervised: 4314 RGB color Infrared thermographic Delamination, splatter, Average accuracy of [13]
delamination and Convolutional neural networks images converted camera images or defect-less 96.8%.
splatters into 18 short
video sequences
Defect prediction of PBF Supervised: Up to 836426 Short wavelength Binary outcome of pore Best prediction [14]
porosities within a 1D CNN model was trained using samples infrared high-speed and non-pore in local accuracy for keyhole
volume Bayesian Optimization camera images volume porosities above
0.1%
Defect prediction of balling, PBF Supervised: 24 385 High speed visible light Prediction of melt pool Accuracy close to 85.1% [16]
under-melting, keyholes, Feature extraction using Scale camera images outcome into 5 types
porosity, spatter Invariant Feature Transforms of desirables
followed by Bag-of-Words conditions, balling,
unsupervised ML then labelling under-melting,
by experts keyholes, porosity,
spatter
(Continued)
www.advmat.de

© 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH

15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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
 Table 2. (Continued)

Research target Fabrication process ML technique Sample size Inputs Outputs Performance Refs.

Process stability in PBF Supervised: 18 000 data points Two IR high speed Prediction of laser focus False positive rate: [17]
predicting laser focus Various ML models ranging from of 9 classes cameras at 700 and 4-class porosity: severe 0.1-0.001%; true
k-nearest neighbours, support (162 000 input 950 nm wavelengths lack of fusion, lack of positive rate:∼90%
vector machines, to vectors) fusion, optimal,

Adv. Mater. 2024, 2310006

convolutional neural network keyhole
Defect prediction in PBF Supervised: Varies with each Visible and NIR high Prediction of keyhole Prediction accuracies [18]
identifying pore Deep neural network configuration up speed camera images defect formation up to 87% were
locations to ≈500 videos achieved
www.advancedsciencenews.com

Defect prediction of PBF Supervised: 1176 time series Acoustic emissions Prediction of keyhole or Best accuracy of 97% [19]
keyhole pore location Support vector machine segments captured via a keyhole-free
microphone fixed 25 formation within a
– 30 cm on top of the time window
build plate
Quality prediction of PBF Unsupervised and Supervised: 22 263 data points Acoustic emissions Part quality (minimum 90% prediction rate [20]
produced coupon K-means clustering captured by a defects or cracks only
Deep learning convolutional wideband sensor at or porosities)
neural network the center bottom of

2310006 (16 of 56)

a circular build plate
Anomaly detection PBF Supervised: 1809 samples Sampling rate: 51.2 kHz Spattering event Highest classification [21]
1D-CNN, 2D-CNN, RNN, LSTM, 512 acoustic signal data confidence of models
Gated Recurrent Unit (GRU) points is 85.08%
Defect detection PBF Supervised: 3000 ML: 23-time domain Three processing Classification accuracy [22]
: Logistic Regression (LR), features, 18 frequency regimes (lack of of ∼93%
Random Forest (RF), and SVM domain features, and fusion
and CNN 263 time-frequency pores, conduction
domain
CNN: rawfeatures
acoustic mode and keyhole
signal with windows pores)
of 5 ms, which is a
time-series signal
consisting of 5000
data points.
Quality prediction of PBF Reinforcement learning 180 spectrograms Optoacoustic fiber Part quality (poor, Prediction accuracies [23]
produced coupon Bragg grating medium, high) above 74%
Defect detection PBF Supervised: >15 000 Four sensors were split Lack of Fusion (LoF), 95.9% to 100% [24]
CNN-LSTM into four different conduction mode,
running windows and keyhole across
(w1, w2, w3, and w4), various time scales
whose time duration
is 0.83, 1.65, 2.5, and
3.30 ms, respectively.
Defect detection PBF Supervised: 4004 Layer-wise imagery, “flaw” and “no flaw” 93.9% and 98.9% [25]
CNN multi-spectral
emissions, and laser
scan vector data
(Continued)
www.advmat.de

© 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH

15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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
 Table 2. (Continued)

Research target Fabrication process ML technique Sample size Inputs Outputs Performance Refs.

Defect detection Material extrusion Supervised: 20 000 value pairs Temperature, humidity, Normal, defect Accuracy: 95% [26]
(FFF) MLP, 1D CNN, RNN LSTM, air pressure, gas

Adv. Mater. 2024, 2310006

Inception time, XceptionTime, particle resistance
XCM
Anomaly detection Material extrusion Supervised: 328 470 Humidity, temperature, Normal, anomaly Accuracy obtained was [27]
(FFF) multi-head encoder–decoder acoustic sensor 97%
www.advancedsciencenews.com

temporal convolutional network

(MH-ED-TCN)
Defect detection Material extrusion Supervised: 120 000 224 × 224-pixel images Over-extrusion, good 98% accuracy [28]
(FFF) CNN-pre-trained ResNet 50 quality,
Under-extrusion
Defect detection Material extrusion Supervised: >8000 512 × 512-pixel images Over-extrusion, good 89.8% [29]
(FFF) CNN- yolov4 quality,
Under-extrusion
Fault diagnosis Material extrusion Semi-supervised: 13 025 images 128 × 128-pixel images standard (STD), low Accuracy for [30]

2310006 (17 of 56)

(FFF) conditional generative adversarial (3.5% for normal nozzle flow rate (Low DANN_CGAN is the
network (CGAN), and domain images, 16% each NFR), high nozzle highest (91.01%)
adversarial neural network for other classes) flow rate (High NFR),
(DANN) low filling speed (Low
FS), high filling speed
(High FS), low
liquefier temperature
(Low LT), and high
liquefier temperature
(High LT).
Real-time defect detection Material extrusion Multi-head deep 946 283 320 × 320 RGB image Good, low, high (for Overall accuracy: 84.3% [31]
and rapid correction (FFF) residual attention network with a flow rate, lateral
single backbone and four speed, Z offset, and
output hot end temperature)
heads, one for each parameter
Detect deformations in the Material extrusion Supervised: 1 M images RGB image from Cropped concrete layers F1 scores for detect and [32]
printed concrete layers CNN model camera non-detect are 82%
Line and edge detections and 98.05%
Thresholding line angles
Detect crack on concrete Material extrusion Supervised: 20 000+ images 300 × 300-pixel images cracks 80% mean average [33]
surface Mobile Net-SSD precision
Detect deformations on Material extrusion Unsupervised: - 3D point clouds Circle radius estimation inaccuracy of 0.3% [34]
printed concrete Principal component analysis and Deformation detection
structures clustering methods (K-means
and spectral cluster)
(Continued)
www.advmat.de

© 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH

15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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
 Table 2. (Continued)

Research target Fabrication process ML technique Sample size Inputs Outputs Performance Refs.

In situ monitoring Material extrusion Supervised: 20 000 images 512 × 512-pixel image 512 × 512-pixel image The pixels in the [35]
Image segmentation using U-net input output segmentized output
neural network have an accuracy of
96.6%,

Adv. Mater. 2024, 2310006

Optimization of printing Material extrusion Supervised: 128 Printing set-up, Extrusion multiplier The results showed an [36]
outcome ad hoc optimized convolutional material type, layer accuracy of 94.3% for
neural network (CNN) and height, infill density overall printing.
mathematical model
www.advancedsciencenews.com

In-process monitoring Material jetting Supervised: A total of 345 20 data points before Droplet size, velocity, Achieve >95% accuracy [37]
(Inkjet printing) Multilayer perceptron-based droplets were used and after each and morphology for droplet
non-linear autoregression local minimum a morphology
(MLP-NARX) model classification
Support vector machine
In situ droplet monitoring Material jetting Supervised: N.A. Droplet size, velocity, Normal dispensing, Accuracy: 90% [38]
(Inkjet printing) Back propagation neural network aspect ratio, non-dispensing, and
presence of satellites satellite modes
In-line process control Material jetting State-space modeling N.A. Nozzle temperature Film thickness Accuracy metric of [39]

2310006 (18 of 56)

(Inkjet printing) 92.3%
In-line monitoring Material jetting Supervised: 150 past datapoints Thickness of printed Thickness of printed - [40]
(Inkjet printing) nonlinear autoregressive neural line line
network with external input
(NARX)
Process stability of laser DED Supervised: 211 images Coaxial infrared camera Four process Prediction accuracies [41]
power and scan speed Deep convolutional neural conditions: normal, above 80%
network low laser power, low
scanning speed, high
scanning speed
Process stability by DED Supervised: 70 112 data points Laser power, scan Melt pool temperature R2 values above 0.51 [42]
prediction of melt pool Extreme gradient boosting speed, layer index,
temperature ensemble time index, average
Long short-term memory neural height, average width
networks
Process stability relating to DED Supervised: 80 000 images Coaxial camera Condition of nozzles (i) Accuracy of 95.9% for [43]
the condition of the K-nearest neighbours normal case, (ii) #1 CNN model
powder nozzles Decision tree nozzle clogged, (iii)
Random forest #2 nozzle clogged,
Convolutional neural network (iv) #3 nozzle
clogged, and (v) #4
nozzle clogged
Realtime monitoring DED Supervised: - Laser power (P), Close-loop feedback - [44]
Artificial Neural Network powder mass flow
(m) and scanning
speed (s)
(Continued)
www.advmat.de

© 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH

15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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
 Table 2. (Continued)

Research target Fabrication process ML technique Sample size Inputs Outputs Performance Refs.

Surface geometrical quality DED Supervised: 247 samples with Laser line scanner Surface conditions: Prediction accuracies [45]
classification RandLA-Net ≈10 000 points normal, convex, above 91.3%
per sample concave
Surface geometrical quality DED Clustering via unsupervised ML 73 samples Laser profiler Surface conditions: no Highest accuracy of [46]

Adv. Mater. 2024, 2310006

classification model followed by several defect, bulging, 93.2%.
supervised learning models: dented, and wavy.
Support Vector Machine,
K-Nearest Neighbours,
www.advancedsciencenews.com

Gaussian Process, Decision

Tree, Naive
Bayes, Artificial Neural Network,
Random Forest, and AdaBoost
Quality of each layer DED Supervised: 400 layers Optical emission Quality of the layer: low Fidelity score from 35% [47]
Support Vector Machine spectrometer and severity, medium to 75%
CCD camera images severity, high severity
Defect prediction (defect DED Supervised: 1300 signal samples Acoustic emissions Acoustic signal with Highest accuracies [48]
free, cracks, or keyhole Random forest, support vector gathered using a removed noise above 89%.

2310006 (19 of 56)

pores) in local area of machine, gradient boosting, microphone Prediction of defect
part and k-nearest, convolutional free, cracks, or
neural network keyhole pores within
the part
Predict surface roughness PBF Supervised: 59 The image texture Surface roughness R2 value of more than [49]
The roughness features such as the parameters 0.9
prediction models are developed contrast from the
using linear regression, GLCM method, and
polynomial regression, support the second moment
vector regression (SVR), from the
Gaussian process regression NGLDM method
(GPR), and artificial neural
networks (ANNs)
Defect inspection (ST and PBF Supervised: – Thermogram sequence Defect detection F-score of 0.96 for large [50]
Inconel) Neural Learning Based Blind detects, 0.57 for
Source Separation (NLBSS) and small defects
Spatial-Temporal Sparse
Dictionary Learning (STSDL)
Algorithm
Detect detection PBF Supervised: Nil 1280 × 1024 Prediction classes: Accuracy >90% for [51]
Image classification via U-net Holes, Spattering, most tested features.
Vertical defects,
Horizontal defects,
and Incandescence.
Pore size prediction PBF Supervised: – Time-series Pore size Classification [52]
Dynamic time warping and temperature accuracies of 92% to
k-Nearest Neighbor classifiers information 94% are achievable
are used on the time-series data
www.advmat.de

© 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH

15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 14. Illustration showing the different types of line morphology created by powder bed fusion using different process parameters and the use of
ML to classify and predict the printability of the parts. Reproduced with permission.[53] 2020, MDPI.

Adv. Mater. 2024, 2310006 2310006 (20 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

was identified, emphasizing the importance of striking an opti- fusing temperature and vibration data from various sensors with
mal balance for future investigations in this domain. It is crucial process parameters. In terms of parts dimensional accuracy pre-
to note that, despite the DL’s model’s substantial speed advan- diction, the residual attention neural network model performed
tage over traditional brute force simulation methods, this does better than other ML models such as 1D CNN and LSTM net-
not imply that the predictions are flawless and devoid of the need works. However, more advancements are required to consider
for further refinement. The model’s success lies in its rapid nar- environmental factors from the outside, develop an online feed-
rowing identification of a singular perfect solution. The model’s back system for real-time prediction, and create a comprehensive
predictions serve as a highly informed starting point, and subse- digital twin system for AM.
quent optimizations can be applied using additional criteria or Bayesian optimization was used in a recent work to accelerate
constraints not fully captured by the training data. This iterative the printability optimization for extrusion-based bioprinting.[57]
refinement process is essential for tailoring the model’s output The input variables for bioink compositions consist of 3-gelatin
to the specific nuances. It is important to note that while the DL methacryloyl (GelMA) concentrations and 3 GelMA/hyaluronic
model significantly outperforms traditional brute force simula- acid methacrylate, whereas the input variables for printer pa-
tion methods in processing speed, this does not imply that the rameters include bioink reservoir temperature, extrusion pres-
predictions are flawless and devoid of the need for further re- sure, print-head speed, and platform temperature. A scoring sys-
finement. The model’s success lies in its rapid narrowing down tem was then implemented to assess the filament morphology
of potential tool paths to a subset likely to include the optimal during extrusion and pore architecture on layer stacking. The
path, rather than guaranteeing the identification of a singular, study has shown that the Bayesian optimization algorithm can
perfect solution. The model’s predictions serve as a highly in- be used to analyze the optimal printer parameters and acceler-
formed starting point, and subsequent optimizations can be ap- ate the extrusion bioprinting experimentation process in com-
plied using additional criteria or constraints not fully captured by parison to the traditional trial and error approach.[57] Another
the training data. This iterative refinement process is essential for work utilized Uniform Design (UD) technique to select 12 experi-
tailoring the model’s output to the specific nuances of any given ment data points based on three parameters four-level data space
LPBF task. Additionally, the well-known challenge of ML models U12 (P3 4 )[58] and SVM algorithm to generate a process map that
not precisely adhering to hard constraints is addressed. In the identified optimal printing parameters to fabricate high-quality
context of LPBF tool path optimization, this limitation necessi- printed parts using Pluronic F127 bioink with a high probability
tates the integration of supplementary optimization algorithms of > 75%.[59] It provided a simple tool to improve the printabil-
or constraint-satisfaction techniques post-prediction. These steps ity of extrusion-based bioprinting process based on width index
ensure that the final tool paths not only approximate the model’s with minimum dataset using inputs such as printing tempera-
predictions but also align with the physical and operational con- ture, material composition, and path height.
straints of the LPBF process. An interesting study used optimized ML models to predict ma-
terial printability for FDM-printed pharmaceutical products.[60] A
total of 318 materials and 1594 formulations obtained from on-
3.2.2. Material Extrusion line literature and in-house formulations were used as dataset
for this study; three different ML techniques (ANN, SVM, and
Multiple process parameters such as the nozzle and bed temper- RF) were used and a 75:25 split was used for training and test-
atures, raster angle, layer thickness, nozzle size, and print speed ing. RF emerged as the best ML model for predicting all targeted
are known to affect the quality of the printed parts during ma- variables (filament mechanical characteristics, extrusion temper-
terial extrusion. The high dimensionality of the dataset warrants ature, printing temperature, and printability) with the highest ac-
the use of ML techniques to identify the most optimum process curacy. Another work compared the optimization of 3D print-
parameters. ing properties for assistive devices using traditional ANN and
A data-driven ML platform using MLP and CNN models deep neural networks (DNN).[61] The DNN outperformed the
was developed to predict optimized parameters for the FFF traditional ANN approach, offering improved calculation speed,
process.[55] Spatial features were first extracted using CNN and higher print quality, and decreased errors. It highlighted the ef-
were then transferred to the MLP model together with other fectiveness of DL-based optimization in 3D printing processes.
process parameters such as extrusion width, layer height, print Another study proposed the use of both open-loop and closed-
speed, infill, area, and volume. The approach enabled quick and loop ML models to monitor the effects of processing param-
accurate predictions of decisive parameters such as time, weight, eters on the quality of 3D-printed parts.[62] The open-loop ap-
and length, even with fuzzy input information. It did not require proach utilized multiple ML classification algorithms such as
consideration of the shape, size, and material of the printed ob- deep neural network (DNN), support vector machine (SVM), de-
ject and can perform the process automatically. The proposed ML cision tree (DT), random forest (RF), and logistic regression (LR)
approach has several advantages, including better stability and to determine the relationship between processing parameters
clearer rules compared to previous research, fast estimation of and printed lines’ quality (large space, little space, good connec-
printer parameters in approximately one second, and applicabil- tion, little material flow, large material flow). A closed-loop sys-
ity to various types of 3D printing materials and domains like con- tem is constructed based on this relationship using a fuzzy in-
struction, medical, and architecture. In a different study, a data- ference system that generates optimized processing parameters.
driven predictive model for the FDM process was created using a The ML-based closed-loop system improved the quality of printed
variety of ML algorithms.[56] The model predicted dimensional parts and enabled a self-adjusting 3D printing process by effec-
deviations between the printed model and the original one by tively monitoring and optimizing processing parameters. More

Adv. Mater. 2024, 2310006 2310006 (21 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

research could be conducted to include additional processing pa- face tension and nozzle diameter.[64] Another study used en-
rameters and conducting real-time closed-loop 3D printing ex- semble learning approach and its base learners (RF, least abso-
periments. lute shrinkage and selection operator (LASSO), extreme gradi-
In larger-scale construction printing, printing parameters ent boosting, and SVR) to predict the droplet velocity and vol-
such as pumping, extrusion, and printing speeds, nozzle diame- ume using inputs such as polymer concentration, excitation volt-
ter, and standoff height have a direct impact on the printing pro- age, dwell time, and rise time in inkjet-based bioprinting pro-
cess and the final mechanical properties of the concrete struc- cess; the experimental results showed extreme gradient boost-
tures. It is a daunting task to identify the optimal printing pro- ing has highest predictive accuracy (R2 = 0.977, RE = 0.044,
cess due to the large number of variables involved. One approach and RMSE = 0.240) in accordance with the studied operating
to address this challenge is through nozzle shape optimization. conditions.[65]
A predictive modeling approach using ANN was proposed to di- Another study utilized ML models to optimize the electrohy-
rectly control the geometry of concrete printing extrudate by opti- drodynamic jet printing of graphene-based biosensors.[66] Su-
mizing nozzle shapes.[63] Thirteen different nozzle shapes were pervised ML models, trained on key printing parameters such
predetermined and used in the experiments, with their corre- as nozzle speed, ink flow rate, and voltage, could predict the
sponding extrudate geometries analyzed using MATLAB. The conductivity of printed circuits in real time. The RF and K-NN
ANN model was then developed to correlate nozzle and extru- (k = 10) models delivered the highest prediction accuracy
date shapes, and a nozzle-extrudate database was formed for an- of about 83%. The integration of ML aimed to streamline
alyzing the optimal nozzle shape for specific target extrudate the manufacturing process, ensure resource efficiency, and
shapes. The results showed a noticeable improvement in sur- produce devices with controlled electrical properties. Over-
face finish quality without additional post-finishing effort, offer- all, the study emphasized the significant potential of ML
ing flexibility for various printing structures with different outer- in enhancing the manufacturing processes in the electronics
surface shapes. The proposed approach has the potential to im- industry.
prove surface finish quality in concrete printing, as it directly ML models were used to predict ink-jetting behavior in the
controls extrudate geometry without the need to reduce nozzle inkjet printing process based on 11 distinct ink and printer
size. parameters.[67] Notably, small ensembles of DT such as boosted
Researchers are making significant advancements in process DTs and RF demonstrated superior predictive power for drop ve-
optimization for 3D printing by exploring nozzle shape optimiza- locity and radius, with an RMSE of 0.39 ms−1 and 2.21 μm, re-
tion, predictive modelling for mechanical properties, and AI- spectively. Furthermore, a neural network model was constructed
based control systems. These approaches offer potential for en- to categorize drop behavior into three categories: stable “sin-
hanced surface finish quality, improved mechanical properties, gle drop”, “multiple drops”, or “no ejection” and achieved an
and better control over the printing process, paving the way for accuracy of 91.94%. The models were validated using an un-
further innovation and application of 3D printing in various in- tried graphene oxide ink, which was not included in the train-
dustries. ing dataset. This innovative ML approach could accurately pre-
dict ink jetting behavior and eliminate the need for costly, time-
consuming, and material-intensive jetting experiments. Overall,
3.2.3. Material Jetting the research demonstrated that ML can significantly enhance the
efficiency of inkjet printing, highlighting its potential for acceler-
There are many variants of jetting-based printing techniques ating the development of new functional ink materials for printed
which include inkjet printing, aerosol jet printing, electrohy- electronics.
drodynamic jet printing, acoustic printing, laser-induced for- An innovative combination of a microfluidics-driven multi-
ward transfer printing, etc., that facilitate drop-on-demand high- scale 3D printer with ML was implemented to enhance the pre-
resolution printing. Each of these printing techniques are de- cision of the freeform generation of active electronics.[68] A new
signed differently and they have their unique sets of process printing and ML workflow was developed to modulate ink com-
parameters that can be controlled to adjust the print qual- position in real time and classify complex internal features. This
ity. Typically, these print parameters are adjusted such that was achieved by using an SVM-guided classification model for
a printed pattern with well-defined edges is obtained. How- automated, in situ pattern classification. The ML model showed
ever, most printing processes have multiple process parame- a balanced accuracy of 81.96% in classifying the internal tex-
ters that make it difficult to identify the most optimum print tures of the evaporative-driven printed droplets. The developed
settings for the best print condition. Thus, many on-going re- ML-integrated printing system facilitated autonomous optimiza-
search works are looking at applying ML to simplify process tion of printing parameters and robust adaptation to unantici-
optimization. pated disturbances. This represented a significant step towards
A multi-objective optimization design method for drop-on- automated process parameter control for the 3D printing of
demand printing parameters through fully connected neural net- electronics.
works (FCNNs) was proposed; a hybrid multi-subgradient de- Another study applied RSM to investigate the interplay be-
scent bundle with an adaptive learning rate algorithm was used tween aerosol jet printing parameters and the intense pulsed
for multi-objective optimization due to its rigorous convergence light (IPL) sintering process for silver nanoparticle film in
theorems; it can be used to optimize printing of droplets with printed electronics applications.[69] The correlation between
smaller diameter, faster droplet speed with absence of satel- print passes and sintering distance on surface morphol-
lite droplets using inputs such as applied voltage, viscosity, sur- ogy and sheet resistance was investigated to elucidate the

Adv. Mater. 2024, 2310006 2310006 (22 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 15. Schematic showing the workflow for a multi-objective optimization using RSM for optimizing the electrical conductivity and surface roughness
in the IPL sintering process. Reproduced with permission.[69] 2022, AccScience Publishing.

complex relationships among the different parameters can be effectively transferred and applied in other printing tech-
(Figure 15). A hybrid multi-objective optimization approach, niques.
including a modified central composite design (CCD) and
non-dominated sorting genetic algorithm (GA), was applied to
systematically manage these conflicting responses. The use of 3.2.4. Directed Energy Deposition (DED)
ML allowed for the identification of optimal windows for the
IPL sintering process, resulting in films with low sheet resis- Besides toolpath optimization for LPBF processes, ML is often
tance and low surface roughness. Compared to conventional applied in other metal printing such as DED and gas-metal arc
trial-and-error methods, this optimization approach was found welding-AM to predict the spatial and temporal thermal fields to
to be more efficient and systematic. This work lays a founda- inform the designers of the producibility and the potential risk
tion for future optimization of IPL sintering parameters for for cracks of the parts.[71] A study utilized a novel approach for
various nanoparticle-based films and multi-layered electronics discretizing the deposition process of gas-metal arc welding-AM
fabrication. process to enhance the adaptability and flexibility of numerical
Various ML methods such as RSM, GA, and transfer learning simulation in analyzing thermal aspects of the material deposi-
were used to optimize the process parameters of the aerosol jet tion process (Figure 16).[71a] A unique data structure was used to
printing process and understand the complexity between their obtain deposition state data from numerical simulation results.
interactions.[70] The results showed that the Gaussian process re- The data was then utilized to train a recurrent neural network and
gression performed better than the other ML models such as K- deep neural network (RNN-DNN), and one convolutional neu-
mean clustering and SVM in terms of prediction of the classifi- ral network (CNN) specifically designed for identifying correla-
cation of printed features such as the line width, edge roughness, tions between deposition stages and their corresponding ther-
and film thickness. It is possible to use a very small dataset for mal fields. The validation results demonstrated that the devel-
the same prediction using transfer learning techniques such as oped method achieved a prediction accuracy exceeding 94% com-
feature representation, instance transfer, and model-based trans- pared to numerical simulation results. Interestingly, the time re-
fer. The feature representation technique outperformed the other quired for a single prediction process was reduced to the millisec-
methods as it resulted in smaller error in the prediction. Overall, ond level. Another study also RNN-DNN for thermal analysis in
the ML works for aerosol jet printing present a framework that laser-aided AM (LAAM).[71b] A thermal field prediction numerical

Adv. Mater. 2024, 2310006 2310006 (23 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 16. Illustration showing the use of RNN-DNN and CNN model to predict the temperature field of the 3D printing process. Reproduced with
permission.[71a] 2021, Elsevier.

model was used to generate a comprehensive training dataset, the the cell-induced light scattering effect.[72] Single-layer trial prints
developed RNN-DNN model successfully correlated laser scan- were obtained from different sample masks and used to train the
ning patterns with their corresponding thermal history distribu- algorithm; a print simulator was used to generate a huge amount
tions. The high prediction accuracy of over 95% compared to fi- of new training data, which greatly reduced the required train-
nite element models highlighted the significance of ML in im- ing samples by more than tenfold. The printed samples and the
proving efficiency and decision-making in LAAM processes. generated samples were then used to train the neural network
This advancement enables rapid evaluation of different scan- which calculates the appropriate masks that compensate the cell
ning patterns within minutes, leading to potential cost savings scattering effect. The NN approach in the study is composed of
and enhanced manufacturing outcomes. Furthermore, the in- two U-Net-like NNs—slave NN and master NN. The network ar-
tegration of ML techniques paves the way for future research chitecture of the master NN is composed of 14 convolution or
on multi-layered 3D deposition processes, expanding the un- deconvolution layers with batch normalization, ReLU, and Tanh
derstanding of complex geometries and optimizing deposition activation function, as well as U-net style skip connections. The
strategies for desired material properties. Overall, these stud- slave NN learned the transformation of the physical 3D printer
ies underscore the importance of ML in revolutionizing thermal and provided gradient information to support the training of the
analysis in metal printing and accelerating its adoption in diverse master NN, while the master NN learned the inverse transforma-
industries. tion of the 3D printer. This allowed the master NN to suggest a de-
formed mask for any given target structure and print it out under
highly scattering condition. Furthermore, the algorithm enabled
3.2.5. Vat Photopolymerization the use of a small sample size (as small as 32) for training data to
generate grayscale masks that can print fine-detailed structures
A recent work has demonstrated the use of NN to learn printing surpassing the traditional manual tuning method with identical
parameters for DLP process from a 3D-printed simulator to offset masks.

Adv. Mater. 2024, 2310006 2310006 (24 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Continuous liquid interface production (CLIP) is an ad-

vanced vat photopolymerization technique that uses an oxygen-
permeable “dead zone” between the fabricated part and trans-
parent window to continually cure the resin. There is an opti-
mal range of printing speeds to achieve successful print for each
defined geometry; a combination of physical modeling and ML
approaches was used in a recent work to identify the optimum
speed and appropriate speed range for continuous printing.[73] A
synthetic dataset was first generated in the absence of an available
experimental dataset to identify the significant factors using the
design of experiments (DOE) for successful prints. The predicted
results for the successful prints were then screened and collected
as new experimental dataset that were subsequently used for
training. Various ML algorithms such as conventional techniques
(DT, naive Bayes, K-NN, and SVM), ensemble approaches (RF,
gradient boost, and Ada boost), and DNN (Siamese networks)
were evaluated and compared; Siamese Networks demonstrated
superior performance (average training accuracy of 90.17% and
testing accuracy of 88.42%) as critical information was extracted Figure 17. Graphical overview illustrating the utilization of ML for opti-
from the mathematical models-generated synthetic dataset mizing designs in the context of AM-related applications.
(Table 3).

the use of different ML techniques for design optimization in AM

3.3. Design Optimization will be provided in the subsequent sections.
A novel approach for constructing AM design rules was in-
The rapid evolution of 3D printing technologies has paved the troduced using ML and knowledge graphs (Figure 18).[75] The
way for innovative approaches in the design and fabrication of framework extracted knowledge on predictive manufacturability
materials and structures. Amidst the plethora of techniques that from data, stored both existing and newfound AM knowledge in
have emerged to enhance the additive manufacturing process, an ontology-based knowledge graph and applied reasoning to de-
the integration of ML stands out as a game-changer. Particularly rive data-driven prescriptive AM design rules. The methodology
in the realm of design and topology optimization, ML offers ca- enhanced the automated and autonomous construction and im-
pabilities that have the potential to redefine the paradigms of provement of AM design rules, supporting AI-related decision-
3D printing. Design and topology optimization traditionally in- making in additive manufacturability analysis and (re-)design for
volve intricate processes, where the objective is to derive the best AM. By providing shareable AM design rule knowledge with the
material distribution within a given space, considering specific AM community, this work promotes collaboration and facilitates
boundary conditions and loads. The challenge here is the vast advancements in the field.
solution space, which becomes computationally intensive and An optimization framework that utilized variational autoen-
time-consuming to navigate. This is where ML comes into play. coder (VAEs) was proposed to design composite mechanical
With its ability to analyze massive datasets, recognize patterns, metamaterials (Figure 19).[76] The focus was on controlling
and make predictions, ML can provide insights and solutions at macroscopic elastic moduli and designing optimal representa-
a pace and precision that are often beyond traditional compu- tive volume element. The approach employed a variational au-
tational methods. Moreover, the iterative nature of design opti- toencoder to learn a reduced representation of representative vol-
mization aligns seamlessly with ML models. These models can ume element configurations, enabling Bayesian optimization for
be trained on a myriad of design variations, learning from each multi-material design problems. Bayesian optimization can be
iteration, and subsequently suggesting optimal design strategies used to construct a probabilistic surrogate model for the objec-
that not only meet but often surpass human-driven solutions. tive function and query the next data point. This ML-based frame-
Furthermore, the incorporation of ML allows for real-time feed- work eliminated the need for subjective trial-and-error design de-
back during the design phase, which can be instrumental in mak- cisions. Experimental validation using multi-material 3D-printed
ing swift, informed decisions. In the context of AM, the choice of samples demonstrated good agreement between the optimized
ML approach is dependent on the stage of design process. Su- values by the ML model and the experimental counterparts.
pervised learning might be useful for the prediction of material A method utilizing StyleGAN was employed to design archi-
properties at the early design stage, while reinforcement learning tected materials inspired by nature, specifically focusing on the
could be employed for fine-tuning and optimizing the design of generative formulation of original unit cell designs inspired by
3D-printed parts. Most of the design optimization problems are leaf microstructures.[77] By employing unsupervised learning,
multi-objective with conflicting design goals. Hence, a combina- this approach facilitates the exploration of a latent space for pi-
tion of approaches, like multi-objective optimization algorithms oneering material design, overcoming the limitations associated
might be applied to determine the optimal solutions. Figure 17 with labelled data. This methodology proves particularly perti-
provides a summary of the application of ML for design optimiza- nent to 3D printing workflows, where it can guide the develop-
tion for various AM-related applications and more discussion on ment of materials and structures by translating natural language

Adv. Mater. 2024, 2310006 2310006 (25 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 Table 3. ML for process optimization in AM.

Research target Fabrication process ML technique Sample size Inputs Outputs Major findings Refs.

Predict printability PBF Supervised: 2048 9 parameter combinations 2 prediction classes (Good A predictive approach for selective laser [53]
(% nano-TiB2 reinforced (SLM) Backpropagation-based of laser power and scan and bad printability) melting was developed, which makes

Adv. Mater. 2024, 2310006

AlSi10Mg composite) neural network model speed use of a ML algorithm capable of
recognizing faulty tracks and
intelligently anticipating printable
parameters
www.advancedsciencenews.com

Toolpath optimization PBF Supervised: 33 000 Laser toolpath Good or bad classification It was noted that the linear model was [54]
(SLS) Linear regression model and based on median not capable of accurately discerning
CNN temperature gradient the true optimal laser path pattern.
Furthermore, when employing a
Convolutional Neural Network (CNN),
the Deep Learning simulation proved
to be significantly faster compared to a
brute force simulation.
Print information Material extrusion (FFF) Supervised: 24640 Extrusion width, layer Time, length, and weight The proposed model does not need to [55]

2310006 (26 of 56)

estimation MLP and CNN height, print speed, infill consider the shape of the object but
percentage, volume area can perform the process automatically
without external factors
Predict dimensional Material extrusion (FFF) Supervised: N.A. Temperature and vibration Dimensional deviations Deep learning algorithms often [56]
deviations random forest (RF), extreme data from multiple outperform traditional machine
gradient boosting (XGB), sensors (thermocouples, learning techniques, even with limited
linear regression (LR), infrared thermometers, datasets. (R2 :0.9113)
gradient boost (GB), light and accelerometers) When tested on a dataset consisting only
gradient boosting machine of process parameters and extruder
(LGBM), decision tree vibrations, the Residual Attention
(DT), Ridge, Lasso, model demonstrated strong resilience
AdaBoost, and three deep and was efficient in predicting the SDM
learning algorithm of the printed components.
models: 1DCNN, The Residual Attention model shows
convolutional neural & improved generalization when exposed
long short-term memory to varied data, and it can accurately
network (CNN-LSTM) forecast the three-layer SDM of
components manufactured using the
FDM process parameters and real-time
sensing information.
Unlike deep learning techniques such as
1DCNN, the attention feature in the
Residual Attention model adjusts
effectively to the dynamic interplays
present in the AM procedure.
(Continued)
www.advmat.de

© 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH

15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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
 Table 3. (Continued)

Research target Fabrication process ML technique Sample size Inputs Outputs Major findings Refs.

Printability optimization Material extrusion Supervised: N.A. Ink composition, reservoir Score system for filament It can be used to analyze the optimal [57]
Bayesian optimization temperature, extrusion morphology printer parameters and accelerate the
framework pressure, print-head extrusion bioprinting experimentation

Adv. Mater. 2024, 2310006

speed and platform process in comparison to the
temperature traditional trial and error approach
Optimization of printing Material extrusion Supervised: 12 Printing temperature, Width index Another work utilized Uniform Design [59]
outcome Support Vector Machine material composition, (UD) technique to select 12
(SVM) path height experiment data points based on three
www.advancedsciencenews.com

parameters four level data space

U12 (P3 4 )[ 58] and Support Vector
Machine algorithm to generate a
process map that identified optimal
printing parameters to fabricate high
quality printed parts using Pluronic
F127 bioink with high probability of >
75%
Printability optimization Material extrusion (Hot Supervised: 1594 samples Material type, glass Filament mechanical Random forest emerged as the best ML [60]

2310006 (27 of 56)

melt extrusion) Artificial neural networks, transition temperature, characteristics, extrusion model for predicting all targeted
support vector machines melting temperature, temperature, printing variables (filament mechanical
and random forests molecular weight temperature and characteristics, extrusion temperature,
printability printing temperature and printability)
with the highest accuracy.
Optimization of print Material extrusion (FFF) Supervised: Maximum tensile force, Material, layer height, Compared with the results from the [61]
parameters ANN, DNN Type of part, Dimensions thicknesses (Top, traditional ANN approach,
of part Bottom Shell), fill optimization based on DL decreased
density, print speed the calculating speed by up to 1.5
Temperatures (bed, 1st times with the same print quality,
and 2nd nozzles) increased quality (both learning:
0.9577 and testing: 0.9721), decreased
MSE (0.001), and a set of printing
parameters not previously determined
by trial and error was also identified.
In situ Process parameter Material extrusion (FFF) Multi-Input-Multi-Output 400 labeled data Error and change in error of Filament extrusion speed, Five supervised ML algorithms – deep [62]
closed-loop feedback (MIMO) Fuzzy logic-based print states Layer height, neural network (DNN), support vector
control control algorithm Line machine (SVM),
Distance, decision tree (DT), random forest (RF),
Print speed and logistic
regression (LR) – are used for
classification. The resulting models are
used to decide the status of printed
line connection given layer height, print
speed, line distance and filament
extrusion speed
(Continued)
www.advmat.de

© 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH

15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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
 Table 3. (Continued)

Research target Fabrication process ML technique Sample size Inputs Outputs Major findings Refs.

Predicting cross section Material extrusion Supervised: 101 extrudate Shape of the nozzle outlet Shape of the extrudate The ML-based proposed approach [63]

Adv. Mater. 2024, 2310006

of extrudate ANN cross-sectional and flow rate cross-section improves the surface quality of three
(concrete) samples were structures with varying curvatures by
generated adjusting the nozzle geometry to
match the desired extrudate shape for
each structure.
www.advancedsciencenews.com

Optimizing Droplet Material jetting (Inkjet Supervised: N.A. Applied voltage, viscosity, Droplet diameter, droplet It can be used to optimize droplet [64]
Formation printing) Fully connected NNs surface tension and speed, satellite droplets printing and propose an optimal range
(FCNNs) nozzle diameter of bio-ink properties for best printing
outcome based on nozzle diameter.
Droplet Prediction Material jetting (Inkjet Supervised: 243 data points Polymer concentration, Droplet velocity, droplet The experimental results showed extreme [65]
printing) Ensemble (random forest, excitation voltage, dwell volume gradient boosting has highest
LASSO, extreme gradient time, rise time predictive accuracy (R2 = 0.977, RE =
boosting, support vector 0.044 and RMSE = 0.240) in
regression) accordance with the studied operating

2310006 (28 of 56)

conditions
Prediction of electrical Material jetting Supervised: A set of 240 Nozzle speed, voltage Conductivity of electrodes Random forest and K-NN (k = 10) [66]
conductivity (Electrohydrodynamic- Random forest, Logistic samples between the print head models resulted in the highest
jet) regression, K-NN and the substrate, and accuracy (about 83%) in estimating the
the flow rate conductivity of graphene electrodes
Prediction of jetting Material jetting (Inkjet Supervised: 3000 data points 11 features (frequency, rise Droplet jetting velocity, Ensembles of DTs (GB and RF) were [67]
window printing) 14 regressive models (linear time, fall time, dwell droplet radius, jettability applied to predict the drop velocity and
regression, Bayes ridge, time, echo time, voltage, prediction, jetting radius of 14 materials. The observed
RANSAC, Decision tree, echo voltage, viscosity, window prediction RMSE was 0.39 m s−1 and 2.12 μm
gradient boosting etc.) surface tension, density, respectively. The mean absolute
and deep neural network nozzle diameter) percentage error is 3.87%. A neural
network model was built to classify
jetting category with 91.94% accuracy.
Tuning of colloidal ink Material jetting Supervised: N.A. Feature maps Deposition patterns SVM model has achieved a balanced [68]
composition and (Microfluidic Support vector machine, accuracy of 81.96% in classifying the
optimization of mixer-based printing) random forest internal textures of the
printing parameters evaporation-driven printed droplets.
Multi-objective Material jetting (Aerosol Supervised: 105 samples Sintering distance, print Electrical resistivity, surface Response Surface Models (RSMs) were [69]
optimization jet printing) Response surface method layers roughness developed, and their associated
(RSM), genetic algorithm statistical uncertainties were
integrated with the NSGA-III algorithm
to effectively optimize the printing
quality in a two-dimensional (2D)
space.
(Continued)
www.advmat.de

© 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH

15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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
 Table 3. (Continued)

Research target Fabrication process ML technique Sample size Inputs Outputs Major findings Refs.

Material jetting (Aerosol Supervised: 100 data points Sheath gas flow rate, carrier Edge roughness, film The RSMs were employed to determine [70a]
jet printing) Response surface method gas flow rate, print speed thickness the optimal operating window in the
(RSM), genetic algorithm 2D space, considering the trade-off
between printed line features and print

Adv. Mater. 2024, 2310006

speed using the desirability function
approach. The derived RSMs and
corresponding statistical uncertainties
were jointly utilized with the NSGA-III
www.advancedsciencenews.com

algorithm to systematically optimize

the overall printing quality in a
three-dimensional (3D) design space.
Material jetting (Aerosol Supervised: 25 samples Sheath gas flow rate, carrier Line width, edge Developed multi-objective optimization [70b]
jet printing) Noisy (input Gaussian gas flow rate, print speed roughness, thickness framework to optimize the overall
process (NIGP)) printed line quality for customized line
width printing using small data set and
prediction uncertainty.
Latin hyper sampling was combined with

2310006 (29 of 56)

a noisy input Gaussian process
(NIGP) approach for rapid modeling
of the printing process
Prediction of features Material jetting (Aerosol Supervised: 90 data points Sheath gas flow rate, carrier Line width, edge It was found that Feature representation [70c]
using transfer learning jet printing) Transfer learning (instance gas flow rate, print speed roughness, thickness transfer generally produce much lower
transfer, feature errors in the prediction of line width,
representation), line thickness and edge roughness.
model-based transfer The errors are found to be 5% or
smaller.
Print quality optimization Material jetting (Aerosol Unsupervised and N.A. Sheath gas flow rate, Edge roughness, thickness, A ML method was introduced to optimize [70d]
jet printing) Supervised: carrier gas flow rate overspray the quality of Additive Jet Printing
K-means clustering, SVM, (AJP) by investigating the connection
GPR, genetic algorithm between the morphology of deposited
droplets and the characteristics of
printed lines. The approach employed
Gaussian process regression to
establish a process model for the
geometrical properties of droplets. The
deposited droplet morphology was
optimized while considering the dual
conflicting objectives of customizing
the droplet diameter and maximizing
droplet thickness. By leveraging this
ML-based approach, the AJP process
could be fine-tuned to achieve desired
droplet characteristics and enhance
the overall printing quality.
(Continued)
www.advmat.de

© 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH

15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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
 Adv. Mater. 2024, 2310006
Table 3. (Continued)

Research target Fabrication process ML technique Sample size Inputs Outputs Major findings Refs.
www.advancedsciencenews.com

Temperature field DED Supervised: - Spatial coordinates, Thermal field, To detect the association between the [71a,b]
prediction (Gas metal arc welding) Recurrent neural network material properties, and Temperature evolution deposition stage and its related
and deep neural network process parameters. thermal field, a physics-based ML
(RNN-DNN) parts, and approach based on an ensemble
one convolutional neural learning model was constructed. The
network (CNN), time cost of a single prediction step
Feedforward Neural was just milliseconds-minutes.
network Accuracy of 94%
Print quality optimization Vat photo-polymerization Supervised: 24 samples Grayscale mask image M of Binary image P of the same The U-net like NNs enabled the use of [72]

2310006 (30 of 56)

(Digital light processing) NN consisting of 2 size 512 × 512 size small sample size (as small as 32) for
U-Net-like NNs (master training data to generate grayscale
NN and slave NN) masks that can print fine-detailed
structures surpassing the traditional
manual tuning method with identical
masks.
Print speed optimization Vat photo-polymerization Siamese networks N.A. Manufacturing speed, resin Printing success/failure Siamese Networks worked the best [73]
(Continuous liquid viscosity, part radius, among all the investigated models
interface production) PDMS thickness, surface (average training accuracy of 90.17%
type, curing time per and testing accuracy of 88.42%) as it
layer, total curing time can effectively extract useful
information from the mathematical
models-generated synthetic dataset
Control the print path of Material Extrusion Deep Reinforcement 5000 simulations Critic network: Trajectory of the toolpath A better performance in terms of process [74]
the aero-pendulum Learning: environment states and Critic network: Expectation duration
extruder (twin-delayed deep agent’s actions of the long-term reward
(concrete) deterministic policy Actor network: Actor network: actions that
gradient) environment states maximize the long-term
reward
www.advmat.de

© 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH

15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 18. Schematic diagram of a framework that utilized ML and knowledge graphs to formalize unstructured AM guidelines into structured knowl-
edge. The framework comprised four key components: a priori knowledge structuration, transformation of AM data into AM knowledge, transformation of
AM knowledge into design for AM ontology, and rule transformation. By employing this approach, the framework facilitated automated and autonomous
construction and enhancement of AM design rules by leveraging both existing knowledge and data-driven insights. Reproduced with permission.[75]
2021, Elsevier.

inputs or human design iterations into optimized 3D models. A technique was developed to reverse engineer composite
This process showcases the potential of integrating advanced ML material parts using imaging methods and ML (Figure 20).[78]
techniques to augment material design and manufacturing pro- The approach captured the geometry of the parts and recon-
cesses, thereby enhancing the efficiency and responsiveness of structed the 3D printing tool path by examining the microstruc-
intricate design workflows to human inputs. ture. The study utilized glass fiber reinforced acrylonitrile

Figure 19. The metamaterial optimization framework involved three main steps. In Step 1, samples were drawn from a random process to create an
artificial database of representative volume element images, each consisting of 28 × 28 pixels. Step 2 involved training a variational autoencoder to
generate realistic output samples. By flipping the 28 × 28 images twice, larger 56 × 56 RVE images were obtained, which preserve symmetry. Finally,
in Step 3, the framework employed Bayesian optimization to achieve the optimal design of a representative volume element that met the specified
macroscopic elastic moduli requirements. Adapted with permission.[76] 2020, Elsevier.

Adv. Mater. 2024, 2310006 2310006 (31 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 20. The implementation of a reverse engineering method using both μCT scan and SEM images of the model. The tool-path details are extracted
by employing an RNN with LSTM architecture, which identifies the fiber orientation within each layer. This approach enabled the reconstruction of the
model’s toolpath based on the available imaging data. Reproduced with permission.[78] 2020, Elsevier.

butadiene styrene filaments for 3D printing specimens, which maximum deformation of the inkjet head by 33%, and lowered
were then reverse-engineered using micro-computed tomogra- the printer’s weight by 19.5%. There was a trade-off between com-
phy (μCT) scans and scanning electron microscopy (SEM) im- puting costs and accuracy, hence it is necessary to discover the
ages. The tool-path information was extracted by identifying fiber optimal balance for future research.
orientation in each layer using an RNN with LSTM architecture. The synergy of ML with design and topology optimization for
The results showed high accuracy in predicting printing orien- 3D printing holds the promise of pushing boundaries, that al-
tation (error of 0.5°) and achieved high dimensional accuracy lows creating efficient and sustainable designs for specific appli-
in the reverse-engineered models. The research demonstrated cations. As we delve deeper into this intersection, we will explore
the potential to reverse engineer high-quality replicas of compos- the mechanisms through which ML augments the design pro-
ite parts by leveraging the capabilities used for designing high- cess, addresses the challenges, and opens the horizons in the dy-
performance composites. namic world of 3D printing (Table 4).
Researchers are advancing the field by integrating ML into AM
processes for automating customization, optimizing parameters,
improving part quality, and facilitating design rule development. 3.4. Microstructure Analysis
These advancements enable efficient and cost-effective produc-
tion, while also promoting collaboration and innovation within The microstructure of metal printed parts is critical for deter-
different research groups. mining their mechanical, thermal, and functional properties.
An optimization method for complex mechanical structures However, there are several challenges in characterizing the mi-
was demonstrated by combining the RSM and multi-objective crostructure in AM parts. The AM process introduces unique
genetic algorithm (MOGA).[79] After conducting experimental complexities such as rapid solidification and cooling rates, which
modal analysis and FEA on the inkjet printer, the method iden- results in distinct microstructural features compared to con-
tified weak points and performance aspects of the structure. The ventional manufacturing methods. Moreover, spatial variations
central composite design (CCD) method was deployed to select within a single part and the need for non-destructive evalu-
sample points for numerical simulations and the initial second- ation further complicate microstructural analysis. To address
order RSM, which focused on the printer’s first-order natural fre- these challenges, advanced imaging techniques such as electron
quency, weight, and maximum deformation of the inkjet head, backscatter diffraction (EBSD), X-ray CT, and high-resolution mi-
was established. The approximation optimization of the RSM was croscopy have emerged as powerful tools for microstructural
then carried out using MOGA, resulting in a Pareto optimal so- characterization in metal printing. These techniques enable the
lution set. The method demonstrated increased computational visualization and quantification of grain morphology, crystallo-
efficiency compared to conventional optimization methods and graphic orientations, and defects. However, analyzing the vast
was good for multi-objective optimization of complex structures amounts of data obtained from these techniques requires so-
printed by the inkjet printer. The optimized solution increased phisticated data analysis methods. This is where ML and DL
the printer’s first-order natural frequency by 36.3%, reduced the algorithms come into play, offering automated approaches to

Adv. Mater. 2024, 2310006 2310006 (32 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 Table 4. ML for design optimization in AM.

Research type Research target Fabrication process ML technique Sample size Inputs Outputs Major findings Refs.

Design ML and KG based - Supervised: N.A. Prior knowledge and Design rule for Data-Knowledge-Design Rule (DKDR) framework, [75]
optimization design rule Classification and surface roughness overhang features which comprises of a prior knowledge structuration,
construction for AM regression tree CART measurement data from AM knowledge to-DfAMOnt transformation, AM

Adv. Mater. 2024, 2310006

a LPBF build data-to-AM knowledge transformation, and Rule
transformation, is a unique framework based on ML
and KG.
Design Automatic design of - Unsupervised: 200 28 × 28 image pattern Elastic moduli Bayesian optimization was used to get the necessary [76]
optimization composite Variational autoencoder (representative volume elastic moduli.
www.advancedsciencenews.com

mechanical (VAEs) element)

metamaterials
Design Toolpath reconstruction Material extrusion Supervised: 78373 images CT-scan images Direction of fiber The original models were reverse engineered with just [78]
optimization using imaging (FFF) Recurrent neural network (70:30) a 0.33% discrepancy in dimensional correctness.
with LSTM
Design Design optimization Material jetting Supervised: N.A. 8 parameters as design Natural frequency, The optimal solution proposed in this study [79]
optimization (Inkjet printing) RSM variables for maximum demonstrates a significant improvement in the
Multi-objective genetic optimization deformation first-order natural frequency of the inkjet printer,

2310006 (33 of 56)

algorithm weight resulting in a remarkable 36.3% increase. This
enhancement effectively mitigates the resonance
region caused by the excitation of the servo motor.
Additionally, the maximum deformation of the inkjet
head is reduced by 33%, leading to enhanced
stability and performance. Furthermore, the weight
of the inkjet printer can be reduced by 19.5%,
resulting in a more lightweight and compact design.
Design Crack prediction PBF Supervised: 540 3D geometries Maximum shear strain Maximum Shear Strain Index (MSSI) as a dependable [80]
optimization High-fidelity surrogate index index for crack formation, it demonstrated how a
model based on an deep convolutional neural network equipped with an
Attention-based U-Net attention-based 3D U-Net architecture could be
architecture trained as a trustworthy surrogate of the
time-consuming construction process simulation,
capable of reliably forecasting MSSI based on the
object’s shape.
Using the surrogate, they applied automated
differentiation to directly compute the gradients of
maximum MSSI relative to the input design
variables. This was then combined with a typical
topology optimization engine’s performance-based
sensitivity field to enable successful design
optimization.
This method considers the trade-off between weight,
manufacturability, and multi-physics performance,
demonstrating a complete approach to dealing with
these numerous factors.
(Continued)
www.advmat.de

© 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH

15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

extract meaningful microstructural descriptors and establish

Refs.

[81]
quantitative relationships between processing parameters and
microstructure variations.
The choice of ML approach for microstructure analysis in AM

Material characteristics ranging from 20A to 65A may

is dependent on the specific objectives and the nature of the mi-

By 3.5%, the ML technology surpasses the surface

crostructure data. Microstructure analysis in AM involves exam-

be simulated using the multi-layer model.

ining the internal structure of 3D-printed parts to understand
their properties and quality. Supervised learning is suitable when
there is labeled microstructure data with corresponding informa-
Major findings

tion about the specific microstructure characteristics or proper-

ties, while unsupervised learning is suitable for discovering pat-
terns, clusters, or anomalies in microstructure data without hav-
response method.

ing pre-defined labels. A combination of supervised and unsuper-

vised techniques may be beneficial for microstructure analysis in
AM; the initial use of unsupervised learning to discover patterns
in the data, followed using supervised learning to predict specific
microstructure based on those patterns. More discussion on the
use of different ML techniques for microstructure analysis in AM
will be provided in the subsequent sections.
216 specimens Infill type, coating shore Shore hardness and

Optimization of AM processes can be performed by leverag-

Outputs

compressive

ing ML models to correlate processing parameters with specific

modulus

microstructural characteristics. A DL framework for the quanti-

tative analysis of microstructural variations in metals fabricated
through additive friction stir deposition technique was developed
to predict the grain size, grain orientation, and grain bound-
hardness, base shore

ary morphology.[82] Microstructural descriptors were extracted

hardness, height,
coating & base

by utilizing EBSD patterns and were used to represent the dif-

Inputs

ferences in microstructures under different processing condi-

thickness

tions. A regeneration neural network was employed to predict

new microstructures within the reduced representation domain.
The framework was validated using samples produced through
parameters)
Sample size

additive friction stir deposition, known for equiaxed microstruc-

combina-
(with 72

tions of

tures. The study addressed challenges in high-dimensional data

design

processing and the identification of principal microstructure de-

scriptors that aligned with specific problem goals. The results
demonstrated the effectiveness of the framework in capturing
salient changes within microstructures and accurately regener-
ML technique

ating them. The physical insights in microstructure descriptors

obtained through mapping the regenerated microstructures pro-
ANN and GE
Supervised:

vided valuable understanding. The study establishes a foundation

for quantifying processing-microstructure linkages in metal AM
and holds promise for applications in materials science, includ-
Fabrication process

ing heterogeneous material design and optimization. Figure 21

provides a summary of the applications of ML in AM for the pre-
Composite layer design Material jetting

diction of various microstructure-related properties.

for tissue-mimicking (Inkjet
anatomical models Printing)

3.4.1. Powder Bed Fusion

Research target

Determining the mechanical properties of 3D-printed metal

parts is crucial for ensuring their reliability, functionality, and
safety. However, there are unique challenges in determining the
mechanical properties of 3D-printed metal parts. The microstruc-
Table 4. (Continued)

ture and defects in the printed material can significantly affect

its properties, making it essential to consider factors like poros-
optimization
Research type

ity, grain structure, and residual stresses. Furthermore, the com-

plex geometry and layer-by-layer fabrication process of 3D print-
Design

ing make it difficult to perform standardized testing. ML tech-

niques offer a promising solution to address the associated chal-

Adv. Mater. 2024, 2310006 2310006 (34 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

dards guide the testing of uniformly prepared specimens that

are assumed to represent the material’s overall characteristics.
The results, such as tensile strength, elongation, compressive
strength, and flexural modulus, are considered to reflect the av-
erage or bulk properties of the material across the entire speci-
men. This assumption is valid under the premise that the spec-
imen is homogeneous, and the material properties are uniform
throughout the specimen. In 3D printing, however, the layer-by-
layer manufacturing process and the potential variability in mi-
crostructure across different regions of a part may challenge this
assumption.
The repeatability of LPBF-printed metal parts was investigated
using ML models (Figure 22a).[90] The mechanical properties
of the printed metal parts (standard deviation of yield strength,
Figure 21. Graphical overview depicting the applications of ML in the field
tensile strength, and maximum elongation) were used to quan-
of AM for predicting various properties related to microstructure.
tify the repeatability. The study showed that the DT method
was the most efficient method to classify and predict the qual-
lenges. By analyzing diverse datasets, ML algorithms can predict ity of the part, achieving an F1 score of 95%. While most ML
mechanical properties based on geometrical and microstructural models are trained with homogenized properties, derived from
features, reducing the need for extensive experimental testing. standardized testing of uniform specimens, their application
These algorithms can also identify correlations between printing extends beyond simple predictions to encompass the complex
parameters and mechanical properties, optimizing the printing and varied nature of arbitrary 3D-printed parts. This potential
process. ML aids in defect detection and classification, enhancing is rooted in the ability of these models to analyze and predict
quality control and it enables more efficient and effective charac- based on diverse inputs, offering a nuanced understanding of
terization of 3D-printed metal parts by leveraging the power of material behavior. For instance, the ML models can be trained
AI. on a dataset that captures a wide range of mechanical proper-
Various ML techniques such as NN,[83] gradient boosting ties, geometries, materials, and print parameters, allowing the
regression,[84] SVM,[85] and GA[86] have been utilized for pre- model to recognize patterns and correlations that apply across
dicting the mechanical property of 3D-printed metal parts with different printing scenarios. By extracting detailed features of
a good coefficient of determination (R2 : 0.84–0.98). The me- an arbitrary part, including its geometry, material composition,
chanical properties that were explored include the ultimate and print settings, ML models could potentially predict its lo-
tensile strength, the maximum elongation, the fatigue life, calised mechanical properties so that designers can rapidly iter-
and the microhardness of the 3D-printed metal parts. The ate on designs by incorporating predictive insights into mechan-
inputs were dependent on the type of mechanical proper- ical properties, effectively tailoring parts to specific performance
ties (static or dynamic). Inputs for static mechanical prop- criteria.
erties include process parameters, parts and build orienta- The adaptation of new printers requires a lot of effort due
tions, surface roughness, relative density, and crystal orienta- to the high variability of 3D printers in determining the opti-
tion, while the inputs for dynamic mechanical properties in- mal process window to fabricate AM parts with good mechanical
clude stress, build orientation, defect size and depth, and loading strength. The main challenge involves the collection of a large
conditions. dataset which can be time consuming and costly. Notably, this
For most cases, the mechanical properties of 3D-printed ma- can be solved using ML models to predict the performance of the
terials are assessed at the specimen level, employing a range of printed parts that are fabricated by a new printer using a small
standardized testing methodologies to ensure consistency, reli- dataset via a transfer learning technique. The published data of
ability, and comparability of results. The American Society for LPBF Ti-6Al-4 V parts was used for the model training to predict
Testing and Materials (ASTM) provides several standards that process parameters for different hardness-porosity property com-
are widely adopted in evaluating the mechanical properties of binations (Figure 22b).[86] The challenges of predicting process-
materials produced by additive manufacturing processes. These property relationship include 1) adopting a new printer model
standards help in defining the procedures for preparing speci- from the same manufacturer, 2. adopting a printer from differ-
mens, conducting tests, and interpreting the results for mate- ent manufacturers but with similar technology, and 3) adopt-
rials such as metals, polymers, and composites. ASTM F2971- ing a printer from a new manufacturer with different technol-
13,[87] ASTM E8/E8M,[88] and ASTM E9[89] are some notable ogy. Bayesian optimization models were found to be effective
ASTM standards used in the testing of 3D-printed materials. in modeling process-property relations and outperformed other
Utilizing these ASTM standards in testing 3D-printed materi- models. The framework demonstrated the feasibility of cross-
als ensures that the mechanical properties are measured ac- machine knowledge transfer and multi-property optimization ex-
curately and consistently, facilitating the comparison of data periments. The work highlights that data mining-assisted ML ef-
across different studies and applications. It also helps in vali- forts can accelerate the development and optimization of metal
dating the performance of 3D-printed parts against traditional 3D printing processes, emphasizing the need for standardized re-
manufacturing methods, aiding in the broader acceptance and porting of data and the creation of a comprehensive metals AM
adoption of additive manufacturing technologies. These stan- database.

Adv. Mater. 2024, 2310006 2310006 (35 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 22. a) Schematic showing the workflow of ML technique for predicting the mechanical property of 3D-printed metal parts. Reproduced with
permission.[90] 2021, Elsevier. b) Schematic showing the workflow for training an ML model for predicting the performance of the 3D-printed parts from
a new printer. Reproduced with permission.[86] 2021, Elsevier.

3.4.2. Material Extrusion perature, room temperature, air gap, and barrel temperature
and showed higher accuracy and lower errors compared to ex-
A study demonstrated prediction of mechanical properties in isting response surface methodology (RSM) models with root-
3D FDM-printed parts. This was addressed by developing a mean-square-error (RMSE) values for ANN and RSM models
data-driven predictive model using LSTM networks for FDM at 0.49 and 0.90, respectively. This approach offers significant
processes.[91] The model took into account the layer-by-layer improvement in predicting part strength, enabling better opti-
printing process and related cyclic layer thermal history to im- mization of manufacturing process conditions and cost-effective
prove accuracy and reliability. Layer-wise activities were captured additive manufacturing. Another study utilized RF and ANN
using sensors (IR sensor, thermocouple, and accelerometer) and (known for their ability to capture nonlinearity) to predict the dy-
their data were incorporated into the LSTM network. The LSTM- namic strength of 3D-printed continuous ramie fiber reinforced
based predictive model outperformed traditional ML techniques biocomposites (CRFRC) under various conditions (varying layer
such as support vector regression (SVR) and RF by 9.8% and thicknesses, hatch spacings, and strain rates).[93] The ANN model
24.3% respectively. Key findings of the study included signifi- outperformed RF in prediction accuracy (5% error compared to
cant improvements in prediction performance by incorporating 9% error) and provided insights into the importance of differ-
in-process sensing data, high relevance of infrared sensor and ent factors. ML proved advantageous in accurately predicting CR-
accelerometer data for tensile strength prediction, and substan- FRC’s dynamic strength, optimizing printing parameters, and
tial contributions of process parameters to tensile strength pre- understanding the influence of microstructural characteristics
diction. The LSTM model demonstrated the effectiveness of se- on composite performance.
quential layer-by-layer modeling of the FDM process, paving the A comparative analysis of selected ML algorithms (ANN, SVM,
way for improved microstructure analysis in AM applications. and RF) was performed for construction printing to evaluate
To address the issue of poor part strength in extrusion-based the interlayer bonding in layered cementitious composites using
additive manufacturing processes, another study developed an non-destructive testing and measurements.[94] The objective was
ANN model to predict the tensile strength of 3D-printed parts.[92] to simplify the mathematical models by reducing the number of
The ANN model considered various input variables such as layer input parameters, making it more practical for implementation
thickness, orientation, raster angle, nozzle temperature, bed tem- in real-world scenarios. The study concluded that ANN yielded

Adv. Mater. 2024, 2310006 2310006 (36 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

the most accurate results in predicting interlayer bond strength,

yielding a linear correlation coefficient R2 of 0.883 and RMSE of
0.341 MPa. The ANN model was trained using data from various
mixtures and it effectively predicted the interlayer bond strength
and determined the optimal printing parameters for obtaining
strong interlayer bonding.
ML models were used to classify 3D-printed boron-based
geopolymer samples based on their compressive strength.[95] Su- Figure 23. A graphical overview summarizing the applications of ML in
pervised ML algorithms such as recursive-partitioning functions material formulation for AM.
rpart and ctree, were used to build separate classification models.
The models were compared in terms of simplicity and cumu-
The choice of ML approach for material formulation in AM is
lative accuracy. The rpart function demonstrated slightly better
dependent on the specific goals and challenges associated with
performance, with a cumulative accuracy of 70% as compared to
the material development process in AM. Material formulation
the ctree function’s 63% accuracy. Furthermore, the rpart func-
in AM involves designing and optimizing the materials with de-
tion required fewer parameters for prediction. The study high-
sired properties for 3D printing. Supervised learning is suitable
lighted the importance of the slag content and the ratio of boron
when there is labeled data that associates material compositions
ions in determining the compressive strength of samples. The
with specific material properties or performance metrics, while
application of machine learning significantly reduces the error in
unsupervised learning is suitable for discovering patterns, clus-
predicting compressive strength, demonstrating its potential for
ters, or similarities among materials or their properties without
developing a guide or standard for classifying 3D-printed boron-
pre-defined labels. A combination of supervised and unsuper-
based geopolymer samples based on compressive strength.
vised techniques may be beneficial for material formulation in
Another study proposed a hybrid approach combining the
AM; the initial use of unsupervised learning to discover complex
multi-objective grasshopper optimization algorithm (MOGOA)
relationship in the data followed by using supervised learning
and ANN to predict the compressive strength of 3D-printed
to build predictive models for specific material properties. More
concrete.[96] The MOGOA was used to optimize the architec-
discussion on the use of different ML techniques for material
ture of the ANN model, considering the number of hidden lay-
formulation in AM will be provided in the subsequent sections
ers and neurons in each layer. The results showed that the hy-
(Figure 23).
brid MOGOA-ANN model achieved accurate predictions (mean
absolute percentage error (MAPE) of 92%) of the compressive
strength, even with simplified neural network architectures. This 3.5.1. Powder Bed Fusion
approach can reduce computational complexity and enable faster
predictions in the material design process (Table 5). Alloy selection profoundly influences the entire AM process,
spanning from the initial energy-source-material interactions to
the final component characteristics. The degree to which lasers
3.5. Material Formulation are reflected or absorbed by a powder bed is contingent upon
the powder’s makeup.[100] Both the internal and external gran-
Material formulation is a pivotal process in achieving desired ular densities of the feedstock contribute significantly to the
properties for various applications. Two key considerations in this density of the end products.[101] Moreover, the thermal proper-
endeavor are processability and the targeted end-use properties. ties of the selected alloy partially dictate the conduction path-
Achieving this balance can be intricate due to the multifaceted ways in the molten state.[102] Variations in solidification rates
nature of material behavior and the interplay between different among alloys can result in significantly diverse post-manufacture
parameters. Processability entails ensuring that the material can microstructures.[103] Certain AM-related challenges, such as the
be effectively and reliably processed into the desired form. On the evaporation of elemental constituents due to intense thermal
other hand, tailoring materials to exhibit specific properties such changes, can be attributed to material composition. This can af-
as mechanical strength, thermal conductivity, or electrical resis- fect the stoichiometry of the melt pools and ultimately the qual-
tivity is crucial for meeting the demands of diverse applications. ity of the finished product.[104] Furthermore, some research has
However, this pursuit is often a delicate balancing act. Enhancing probed the influence of feedstock attributes, like particle size dis-
one aspect may inadvertently impact another. For instance, in- tribution and shape, on process outcomes.[100b,105] Yet, the exact
creasing mechanical strength might result in reduced flexibility. effects remain to be fully elucidated.
Moreover, this trade-off is complicated by the multitude of factors Databases offering a myriad of alloy attributes serve as indis-
at play, including chemical composition, processing conditions, pensable tools in the realm of materials science. Notably, the
and material microstructure. This is where ML steps in as a pow- International Crystal Structure Database (ICSD) hosts crystallo-
erful tool. By ingesting and analyzing vast datasets encompassing graphic structures of countless materials, while the Linus Paul-
various material compositions and their corresponding proper- ing files extend from atomic specifics, such as radii and electron
ties, ML algorithms can identify complex relationships and pat- valence, to more advanced crystallographic information.[106] Con-
terns that might elude traditional analysis. This enables the for- temporary platforms like Aflow[107] and the Materials Project[108]
mulation of predictive models that guide the selection of optimal empower users with sophisticated search functionalities across
material compositions to achieve desired properties, while also diverse alloy datasets. The efficacy of data mining in advanc-
considering processability constraints. ing AM alloy development has been underscored by several

Adv. Mater. 2024, 2310006 2310006 (37 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 Table 5. ML for microstructure analysis in AM.

Research target Fabrication ML technique Sample size Inputs Outputs Performance Refs.

Adv. Mater. 2024, 2310006

process

Predict Additive friction Supervised: Nil EBSD scans Grain size, Near 100% accuracy [82]
microstructure stir deposition convolutional neural grain orientation, and grain
network (VGG16) boundary morphology
www.advancedsciencenews.com

Predict mechanical PBF Supervised: 7680 Crystal orientation as input Mechanical properties R2 value of 0.84 and a root mean [83a]
property a custom 3D (with or without auxiliary square error (RMSE) value of
convolutional neural input features) 16.57 MPa.
network (CNN) based
on VGGNet
Prediction of fatigue PBF Supervised: 32800 data points Stress, build orientation, Fatigue life R2 value of greater than 0.98 [83b]
life Backpropagation neural defect size, defect depth,
network defect distance to
surface

2310006 (38 of 56)

Mechanical property PBF Supervised: 3000 Surface Mechanical property R2 value of greater than 0.98 [84]
prediction Gradient boosting roughness and void
regression position, number density
and size
Fatigue life PBF Supervised: - Process parameters, Fatigue life R2 value of greater than 0.95 for [85]
ANN, random forest, fatigue loading random forest and SVM
Support vector conditions
machine
Transfer learning for PBF Supervised: Data from 120 Laser power, scan speed, Relative density, Overall prediction accuracy of [86]
property Multi-Objective Genetic manuscripts energy density, hatch microhardness, 88%
prediction Algorithm (MOGA) spacing, layer thickness,
powder size
Repeatability of PBF Supervised: 167 Production parameters Static mechanical F1 score > 0.95 [90]
printed parts Tree-based models such as build mass, properties such as yield
height and time, part strength, tensile
volume, sample strength, and maximum
locations and elongation.
temperature parameters.
Tensile strength Material extrusion Supervised: Material property, extruder Tensile strength With an RMSE of roughly 2%, the [91]
prediction (FFF) LSTM temp, printing speed, developed LSTM model
layer height outperforms previous ML
models in predicting the tensile
strength of 3D-printed parts.
(Continued)
www.advmat.de

© 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH

15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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
 Table 5. (Continued)

Research target Fabrication ML technique Sample size Inputs Outputs Performance Refs.
process

Part strength Material extrusion Supervised: 120 (70:15:15) Layer thickness, Tensile strength Pellet based printing: The [92]
prediction (FFF) ANN, RSM orientation, raster angle, RMSE for the ANN results are
raster width, air gap lower in most of the cases,

Adv. Mater. 2024, 2310006

which shows the adequacy of
this model, but ANN is not
suitable for less data, while
RMSE for the RSM sometimes
www.advancedsciencenews.com

performs better in cases with

less data using thermal imaging
to predict the strength.
Dynamic tensile Material extrusion Supervised: 108 (81:27) Hatching spacing, layer Tensile strength The workflow combines [93]
strength (FFF) Random forest, ANN thickness, strain rate dual-resolution Robotic
prediction Scanning, Neural Network
prediction and printing of PETG
plastic.
This integrated approach offers

2310006 (39 of 56)

the advantage of responding
directly to unknown geometries
through automated
performance design
customization.
Interlayer bond Material extrusion Supervised: 338 Parameters of concrete Pull-off adhesion ANN provided the most accurate [94]
strength prediction ANN, SVM, RF surface repair cover assessment of the interlayer
(concrete) thickness: 10-point bonding pull-off adhesion of
height, core height, layered concrete, with an
reduced peak height, average relative error of 10.13
average mobility, overlay percent.
thickness,
frequency of the sound
wave
reflection from the bottom
of
a sample
Compressive Material extrusion Supervised: 114 targets Weight percentages of the Compressive strength Ctree function has a 100% [95]
Strength The conditional inference fly ash and positive predictive value, while
prediction trees (ctree) and Ground granulated rpart has a positive predictive
(concrete) recursive partitioning blast-furnace slag value of up to 81%.
(rpart) methods (GGBFS) (%S), as well
as the ratios of boron
ions
silicon ions and the ratio of
sodium
ions in the alkaline solution
(Continued)
www.advmat.de

© 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH

15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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
 Adv. Mater. 2024, 2310006
Table 5. (Continued)

Research target Fabrication ML technique Sample size Inputs Outputs Performance Refs.
www.advancedsciencenews.com

process

Compressive Material extrusion Supervised: N.A. Water-cement Compressive strength Accuracy of ANNMOGOA-1 is [96]
strength Multi-objective ratio (W/C), amount of about 92%
prediction Grasshopper coarse aggregate (CA),
(concrete) Optimization amount of fine aggregate
algorithm’s (FA), amount of
hybridization with ANN super-plasticizer (S)
(ANNMOGOA)
Surrogate model to Material extrusion Supervised: 1800 Number of TPMS layers Peak load The maximum deviations of 2.5% [97]

2310006 (40 of 56)

predict peak and ANN and volume fraction Maximum displacement and 3.5% for peak loads and
maximum maximum midpoint
displacement of displacements
beam of cellular
design (concrete)
Predicting the Material extrusion Supervised: 850 combinations Design parameters: aspect Poisson’s ratio R2 > 0.99 and MSE < 0.0004 [98]
auxetic behavior perceptron-based neural FEA simulation ratio, length of the major
of cementitious network (NN) results axis, number of voids
cellular Shapley additive along the X- and Y-
composite explanations (SHAP) directions
(concrete)
Crack Material extrusion Supervised: 193989 (95:5) 2D microstructure image Crack patterns CNN model predictions show [99]
prediction of U-net CNN containing pores (32 by excellent agreement with lattice
air-void structure 32 pixels) numerical analyses, with an
(concrete) Intersection over Union (IoU)
of 0.85 for crack pattern
prediction and an R2 value of
0.75 for stress-crack width curve
prediction.
www.advmat.de

© 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH

15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

studies. A study employed this method to optimize the chemistry perimental data (R2 of 0.998 for yield stress and R2 of 0.978 for
of aluminum alloys for enhanced processing during LPBF.[102] plastic viscosity). The models demonstrated high efficiency and
The printed Al alloys historically faced challenges such as lim- predictability, with performance index factors indicating their
ited grain nucleation, resulting in the coalescence of expansive accuracy. Various statistical parameters and external validation
grains and concomitant intergranular stresses, predisposing the checks further confirmed the precision and generalization capac-
material to hot-cracking. To mitigate this, a study sought po- ity of the GEP models in predicting the rheological properties of
tential grain inoculant compounds, generated via chemical re- fresh concrete.
actions during LPBF, which could refine grain properties. A Another study focused on predicting the static yield stress
notable solution emerged when silicon and carbon reactions (𝜏Sn ) of blended cement pastes containing supplementary ce-
birthed SiC particles, promoting more uniform grain nucleation. mentitious materials (SCMs) using ML models.[112] A dataset
However, the lattice disparities between certain compounds and from previous experimental work was collected and eight input
the aluminum alloy could induce substantial stresses at their parameters, including SCM properties, cement reactivity, mix-
interface, perpetuating cracking issues. Consequently, the fo- ture design parameters, and resting time were identified. A com-
cus shifted to identifying inoculants with lattice parameters parison of different ML models (MLP, RF, and SVR) was con-
aligning with the primary aluminum alloy. Through the de- ducted and the MLP model demonstrated the highest accuracy
ployment of a search algorithm probing around 4500 poten- with low RMSE and high coefficient of determination (R2 ). The
tial nucleants, hydrogen-stabilized Zr emerged as the optimal study revealed that 𝜏Sn was primarily influenced by the amount
candidate. of pseudo-contact points, while the amount of cement replace-
The challenges of data mining from existing literature in- ment by SCM had the least effect by analyzing the importance of
clude missing values from isolated studies of individual pa- different input parameters using Shapley-value and permutation
rameters or properties. A database linking process parameters feature importance analysis. These ML models show promise in
and material properties in SLM-fabricated Ti6Al4V alloys was improving mix design for innovative concrete technologies that
developed.[109] Various data imputation methods such as K-NN, require better workability control such as concrete printing or
multivariate imputation by chained equations, and graph im- self-consolidating concrete.
putation neural network (GINN) were explored to fill missing Apart from the rheological properties, ML has also been used
data. The K-NN model excelled in process parameters, whereas to predict the optimal composition of additives for the feedstock
GINN model excelled in material properties. The imputation material. A study investigated the 3D-printing of polylactic acid
quality was enhanced by using the median of the values from the (PLA) composites reinforced with chopped long carbon fiber us-
three models, and a self-organizing map provided visualization ing an FDM printer.[113] The study employed gaussian process
of the relationships between process parameters and material modeling, an ML technique suitable for small dataset, to predict
properties. the optimal carbon fiber content for the composites The model
predicted the best mechanical performance at 6.7 wt.% carbon
fiber, closely aligning with the experimental result of 5 wt.% car-
3.5.2. Material Extrusion bon fiber. The use of ML demonstrated its advantages in accu-
rately predicting material properties and optimizing composite
One key area of focus in material formulation for material extru- performance, potentially saving time and resources in the man-
sion technique is the rheological properties of the extrudate to ufacturing process.
ensure good flowability and extrudability. A study was conducted
to investigate the effects of various admixtures on the rheologi-
cal properties of cement paste for 3D printing applications.[110] 3.5.3. Material Jetting
An empirical formula was proposed to analyze the relationship
between dynamic yield stress and mini-slump. ANN model was Material design plays a crucial role in 3D printing of electronics,
used to predict dynamic yield stress and mini slump based on ad- as it directly influences printability and the final properties of the
mixture proportions. The model was validated by simulating new printed traces (conducting, semi-conducting, or insulating). Ad-
mixes, and the results demonstrated a high level of effectiveness ditives such as surfactants or binders are typically added to the
in predicting the correlation between mini-slump and dynamic functional materials to ensure good printability of the functional
yield stress. This work opens avenues for future research to con- inks. The proportion of different solvents is important in deter-
sider the time factor in ANN models for predicting printability mining the printability and the final deposition pattern for cer-
over time. tain inks. Important factors such as particle loading, and particle
A novel ML algorithm called gene expression programming size can influence the printability and final electrical property of
(GEP) was used to develop mathematical models that predict the the printed structures. It can be a daunting task in identifying
rheological properties of concrete such as yield stress and plas- the optimal parameters and ratios for the materials. Therefore,
tic viscosity.[111] A comprehensive database was built using pre- researchers are increasingly turning to ML techniques to predict
vious experimental results and the significant input parameters the performance of the final printed electrical circuits and com-
that influence concrete rheology (cement, sand, water, small and ponents, streamlining the process, and improving overall out-
medium-sized coarse gravels, and superplasticizer) were identi- comes.
fied. The GEP models, which use simple arithmetic expressions A study introduced a strategy that leveraged ML models to
to describe the relationships between the input parameters and guide the 3D printing process of Cu anode scaffolds directly
the rheological properties, exhibited a strong correlation with ex- onto solid-state NASICON-type Li1+x Alx 3+ M2−x 4+ (PO4 )3 (LATP)

Adv. Mater. 2024, 2310006 2310006 (41 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 24. Illustration showing the ink formulation optimization using RSM to improve the electrical resistivity and line quality. Reproduced with
permission.[115] 2023, Nature Portfolio.

electrolytes, thereby addressing obstacles associated with lithium 4. Outlook

batteries.[114] A sequential learning method based on mixture
design was employed to refine the formulation of the printing 4.1. Advanced ML Models
ink, the rheological parameters, and the operational conditions of
the 3D printing process. The method streamlined experimental Large Language Models (LLMs) are revolutionizing the ML land-
work by systematically enhancing the design variables through scape, showcasing broad applicability across domains due to
a sequence of experiments, resulting in a sturdy methodology their profound understanding of language and context. These ad-
for ink characterization. The printed hierarchical scaffold and vanced AI systems are designed to comprehend, generate, and
copper oxide (CuO) interlayer substantially decreased overpo- interact with human language at scale, excelling in tasks like
tential in comparison to unadorned lithium anodes, which was writing, translating, summarizing, and question-answering. Fu-
achieved by improving interfacial contact to mitigate the forma- elled by DL techniques and the transformer architecture, ex-
tion of lithium dendrites and inhibit side reactions. In summary, emplified by OpenAI’s GPT series, LLMs process and generate
the research introduced a novel methodology for swiftly explor- coherent, contextually relevant text. “MechGPT”, developed by
ing anode geometries and 3D printable inks for lithium batter- Buehler’s group, exemplifies their prowess in modeling mechan-
ies, exploiting both experimental design and machine learning ics and materials, showcasing the model’s proficiency in knowl-
techniques. edge retrieval, hypothesis generation, and bridging disparate ar-
A hybrid multi-objective optimization technique was used to eas for understanding and predicting material behavior and fail-
identify the best functional ink composition for aerosol jet print- ure mechanisms.[116]
ing technology and obtain low electrical resistivity and good Recent efforts have also focused on enhancing the efficiency of
printed line quality (Figure 24).[115] The suggested method sys- in-context learning for LLMs through active learning strategies,
tematically examined the causal relationship between several ink specifically targeting the optimization of demonstration selection
components (ethanol, nanoparticle silver ink, and carbon nan- for few-shot learning tasks.[117] By employing methods such as
otube (CNT) ink) and printing outcomes. Two RSM were created uncertainty sampling, diversity sampling, and similarity-based
based on the analysis of variance; a non-dominated sorting ge- selection, the study identifies the most informative examples that
netic algorithm III (NSGA-III) was then merged with these mod- significantly improve LLM performance. The principles from this
els to provide a more reliable optimization in the 3D mixture de- research can be adapted to 3D printing in several ways – 1. opti-
sign space. This data-driven methodology extended the process mizing printing parameters through active learning to enhance
of creating materials with multi-component and multi-property the quality of 3D-printed objects while minimizing resource use,
in aerosol jet printing technology, resulting in higher electrical 2. accelerating the design iteration process through active learn-
performance and broader applications in the field of printed elec- ing to facilitate rapid prototyping of models based on feedback
tronics (Table 6). and predicted outcomes, 3. developing new printable materials

Adv. Mater. 2024, 2310006 2310006 (42 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 Table 6. ML for material formulation in AM.

Research target Fabrication ML technique Sample size Inputs Outputs Major findings Refs.
process

Rheological Material Supervised: 16 Dosage of high efficiency Dynamic yield The suggested model can precisely forecast the [110]
properties extrusion ANN water reducing agent, stress, mini slump dynamic yield stress and mini-slump of

Adv. Mater. 2024, 2310006

prediction hydrated calcium silicate, 3D-printed cementitious materials using a
(concrete) nano-clay, viscosity-modified variety of admixtures of different kinds and
agents, accelerating agent quality.
Rheological Material Supervised: 137 data sets Percentages of Cement, water, Yield stress and For yield stress and plastic viscosity, the [111]
www.advancedsciencenews.com

parameters extrusion Gene expression of yield sand, small coarse gravel, plastic viscosity mathematical models based on GEP
prediction programming stress and medium coarse gravel, and demonstrated greater efficiency with
(concrete) (GEP) 142 for super plasticizer significant correlation factors R2 of 0.978 and
plastic 0.998, respectively, with the experimental data
viscosity (R2 ∼1).
Yield stress Material Supervised: 280 data points SCM properties (i.e., particle Static yield stress The best model to predict 𝜏S0 was MLP, [112]
evolution extrusion ANN, Random (75:25) number density (NPSD), which had an average RMSE of 20.97 Pa and an
prediction Forest, SVR specific surface area R2 of 98.4%.
(concrete) (SSAPSD), surface potential

2310006 (43 of 56)

(𝜁), and
hydraulicity (SCMR)), cement
reactivity (CR), main mixture
parameters, and resting
time (t)
Mechanical Material Supervised: - Composition of CF/PLA Tensile strength, It can accurately gauge uncertainty metrics and [113]
properties extrusion Gaussian Process tensile modulus, offer a distribution for the predicted value.
prediction Modeling elongation at break,
flexural strength,
flexural modulus,
hardness
Optimization of Material Supervised: N.A. Solid loading, binder Herschel-Bulkley Successfully identified the optimal range of ink [114]
ink formulation extrusion Mixture design humectant shear thinning formulation, referred to as the printable
(Direct ink RSM, sequential index, dynamic window, within which the desired printing
writing – learning yield stress, static results were achieved. The optimal
pneumatic) yield stress, ink rheological parameters for the selected ink
stiffness G’ LVR, deviated from the initial projections and
line continuity, % previous findings reported in the literature
spreading
Optimization of Material jetting Supervised: 13 data points Ink composition of silver, CNT, Resistivity, line quality Investigated the causal relationship between [115]
ink composition (Aerosol jet Mixture design and ethanol various ink components, namely nanoparticle
printing) RSM, silver ink, CNTs ink, and ethanol, and their
non-dominated corresponding effects on the printing
sorting GA outcomes. To optimize the composition of
functional ink to achieve desirable
characteristics, specifically low electrical
resistivity, and high quality of printed lines.
www.advmat.de

© 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH

15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

through active learning to prioritize the exploration of material puted tomography,[124] and infrared signal-based sensors like
compositions to yield novel properties and expedite innovation pyrometers[125] and high-speed infrared cameras, focus on differ-
in material science within the 3D printing domain. ent aspects of monitoring. Notably, multi-sensor signal integra-
Moreover, LLMs, with their vast knowledge base, prove valu- tion, combining accelerometers, acoustic emission sensors, and
able in troubleshooting 3D printing issues by offering language- optical emission spectrometers with CCD cameras, enhances si-
based solutions. They can understand descriptions of issues en- multaneous monitoring.[126] Employing ML models alongside
countered during the 3D printing process and provide relevant the various existing sensors aids in detecting macroscale or
solutions or suggestions for parameter adjustments. Their ca- mesoscale defects. However, a trade-off exists between the speed
pacity to learn from diverse data sources enables them to offer of processing 1D data and higher information density for 2D or
guidance on a wide range of problems, from hardware malfunc- 3D data, posing a challenge in balancing processing speed and
tions to software glitches and material issues, making them valu- information richness in AM defect detection strategies.
able tools for both novice and experienced users seeking to opti- In the dynamic realm of AM, the integration of advanced sen-
mize their 3D printing workflows. For instance, ChatGPT was ex- sor technology with ML models presents a transformative oppor-
plored to optimize the G-code generation process in AM, partic- tunity to push the boundaries of innovation. Looking ahead, the
ularly focusing on fused filament fabrication (FFF) with thermo- adoption of three-dimensional data acquisition, particularly us-
plastic polyurethane (TPU) as the feedstock.[118] It assesses Chat- ing stereoscopic cameras, emerges as a significant advancement
GPT’s capabilities in addressing common 3D printing challenges over traditional two-dimensional imaging methods.[127] These
such as warping, bed detachment, and stringing by optimizing cameras capture the manufacturing process in three dimensions,
printing parameters. The study demonstrates ChatGPT’s effec- unlocking a previously inaccessible depth of data. This enhance-
tiveness in generating optimized G-code, leading to improved ment allows ML algorithms to predict the quality and perfor-
print quality and material savings. It highlights the potential of mance of the final product more precisely, shifting from a flat per-
integrating AI tools like ChatGPT in additive manufacturing to spective to a comprehensive, volumetric analysis that could revo-
enhance efficiency, reduce trial and error, and accelerate innova- lutionize quality assessment practices. Expanding the data spec-
tion in material science. trum for ML models, the incorporation of advanced measuring
Incorporating generative AI into design workflows holds techniques such as scanning electron microscopy (SEM),[128] sur-
promise for transforming 3D modeling and printing. This vi- face roughness measurement,[129] and in situ computed tomog-
sion entails a collaborative experience where designers articu- raphy (CT) scans provide abundant information on both micro-
late ideas in natural language, and AI translates them into tan- and macro-scales.[130] SEM offers insights into microstructural
gible 3D models. The iterative cycle of feedback and refinement integrity crucial for predicting mechanical properties, surface
between human designers and AI could significantly accelerate roughness serves as a direct quality metric, and CT scanning
the design process, allowing for rapid prototyping and optimiza- non-destructively verifies internal structure and dimensional ac-
tion, thereby blurring the lines between imagination and mate- curacy. The convergence of these diverse data streams holds the
rialization in the realm of 3D printing. A method was proposed potential to create a comprehensive dataset for ML models, capa-
for generating 3D architected materials from natural language ble of transforming process optimization, and real-time quality
inputs using a combination of a vector quantized generative ad- control in AM.
versarial network (VQGAN) and contrastive language-image pre- The concept of data fusion is particularly promising in this
training (CLIP) neural networks.[119] This approach translates context, holding the potential to establish robust, predictive ML
natural language descriptions into 2D images, which are then models providing comprehensive insights into the AM process.
converted into 3D models for printing, applying both to mate- By amalgamating data from various sources, such as stereoscopic
rials with varying rigidity and to molecular dynamics modeling images, SEM analyses, surface topology, and CT scans, a more
of nano-architectures. This innovative method allows for the di- nuanced view of the printing process emerges. ML models can
rect materialization of concepts derived from language, offering be trained on datasets reflecting the complexity of AM, enabling
new pathways for complex design workflows in 3D printing by more accurate predictions and the ability to proactively iden-
leveraging human-readable inputs to drive the creation and opti- tify and address potential failures. Furthermore, the use of aug-
mization of 3D models and materials. Despite their capabilities, mented and virtual reality (AR/VR) technologies can potentially
these models face challenges such as bias, interpretability, and enhance this advanced sensory ecosystem, offering an immersive
adaptability in novel situations. interface for design and decision-making.[131] When integrated
with ML, AR/VR creates a virtual testing ground for refining de-
signs and simulating manufacturing outcomes, pre-empting po-
4.2. Advanced Sensors tential issues and optimizing parameters for optimal results.[132]
Informed by both virtual simulations and real-time sensory data,
Various sensors, ranging from image-based to sensor signal- ML models guide users through the design-to-production jour-
based types, are utilized to monitor and detect defects in ney, suggesting modifications that enhance the end product’s
AM, providing comprehensive insights into the processes.[120] functionality and design fidelity.
Image-based sensors, capturing visual and sequential images The integration of these technologies into AM not only ele-
through cameras, offer a detailed representation of the print- vates the precision and reliability of the manufacturing process
ing process.[121] Sensor signal-based techniques, including acous- but also represents a stride toward a fully integrated, intelligent
tic emission with fibre Bragg grating (FBG) sensors,[122] op- manufacturing system. These advancements signal the potential
tical emission with multispectral sensors[123] and X-ray com- for a new era of “smart” AM, where machines evolve beyond tools

Adv. Mater. 2024, 2310006 2310006 (44 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

of creation to become design partners capable of learning, adapt- demonstrated the use of ML approaches to predict cell viability
ing, and optimizing in real-time, ensuring the delivery of prod- during the bioprinting processes with high accuracies.[143] Vari-
ucts meeting the highest standards of quality and performance. ous parameters were evaluated in the extrusion-based bioprint-
ing system[143a] and vat photo-polymerization bioprinting,[143b]
and the live-dead assays provided the dataset for ML training
4.3. ML Applications in Emerging AM-Related Fields to predict cell viability post-printing. Additionally, a recent study
demonstrated the ability to predict the number of printed cells
In the vibrant landscape of AM, 3D printing has taken center in inkjet-based bioprinting based on the droplet velocity profile
stage, metamorphosing various industries by offering unprece- captured using a high-speed camera.[144] The ability to precisely
dented customization, precision, and efficiency. As this trans- predict the number of printed cells is important for fabrication
formative technology continues to evolve, the integration of ma- of 3D tissue constructs in a scalable and reproducible manner.
chine learning emerges as a compelling frontier, promising to Finally, the tissue maturation process plays a vital role in cell
amplify the capabilities of 3D printing across diverse AM-related proliferation and differentiation over time, ultimately resulting in
research areas. In this section, the unique applications of ML in 3D tissues/organs with some degree of functionality. This intri-
various fields such as metal printing, polymer printing, bioprint- cate process involves critical biochemical cues that regulate cellu-
ing, construction printing, drug printing, and electronic printing lar behavior within 3D-bioprinted tissue constructs. Biomechan-
will be discussed. ical conditioning, including mechanical conditioning,[145] elec-
tromechanical stimulation,[146] macromolecules,[147] air-liquid
interface cultivation,[148] or short-term hypoxic conditions,[149]
4.3.1. Bioprinting has been explored to expedite the maturation of 3D-bioprinted
constructs into vascularized, functional tissues. ML approaches
Over the years, the widespread adoption of 3D bioprinting tech- have been applied to assess the differentiation potential of
nologies (including extrusion-based,[133] jetting-based,[134] vat cells using morphological-based prediction by measuring gene-
photopolymerization-based)[135] in tissue engineering, regener- expression profiles and various biomarkers of undifferentiated
ative medicine, and biomedical applications can be attributed cells.[150] Furthermore, a recent study utilized mineral apposition
to its remarkable ability to accurately deposit multiple types of rate and mineralizing surface area as input loading parameters
living cells and bio-inks at pre-defined positions. This capabil- in a DL model to predict and accelerate loading-induced osteo-
ity facilitates the fabrication of biomimetic 3D tissue-engineered genesis during the bone remodeling process.[151] The interplay
constructs.[136] The pivotal functional units in 3D bioprinted tis- between advanced ML models with detailed biological parame-
sue constructs are the living cells, and ML proves to be a po- ters promises to revolutionize the ability to predict and influence
tent tool for unraveling the complexities of cellular behavior. It tissue maturation, marking a significant stride toward the real-
achieves this by handling large datasets, identifying patterns, and ization of functional, bioprinted tissues.
making predictions at various stages of the bioprinting process
(Figure 25).
Cell expansion is a crucial step in the bioprinting process to at- 4.3.2. Bioelectronics
tain sufficient and well-characterized cell populations for the de-
velopment of functional and viable tissue constructs. A study im- The convergence of bioprinting and electronics printing in bio-
plemented an innovative ML approach, employing just-in-time electronics printing marks a revolutionary development, ush-
learning to calibrate Raman spectroscopic models. This enabled ering in a new era of advancements in healthcare and related
real-time predictions of critical cell culture performance param- domains.[152] This technology fundamentally revolves around
eters for optimal cell growth.[137] ML has also been applied to creating a platform that combines biomaterials with electronic
differentiate healthy from apoptotic cells based on cell size and components, paving the way for a spectrum of devices, from
granularity information.[138] Flow cytometry-based analysis of cell portable benchtop platforms to wearable or implantable plat-
size and granularity, combined with ML, offers an automated, re- forms. These devices seamlessly interact with biological systems,
liable, and stain-free classification of healthy and apoptotic cells. holding transformative potential in areas such as regenerative
ML plays a pivotal role in optimizing cell proliferation and se- medicine, neural interfaces, biosensors, and in-vitro diagnos-
lecting healthy cells for the fabrication of 3D biomimetic tissue tic tools for medical assessments and drug testing. The advent
constructs. of 3D multi-material printing technology has enabled the fu-
Numerous publications discuss the use of ML for optimiz- sion of diverse materials, spanning biomaterials to functional
ing printability in bioprinting[36,139] and a comprehensive under- electronic inks. This synergy provides unparalleled integration,
standing of the influence of printing parameters is essential to design flexibility, and functionalities that traditional platforms
enhance the viability of the printed cells. Various factors such struggle to achieve. Nevertheless, bioelectronics printing encoun-
as shear stress, nozzle size in extrusion-based bioprinting,[140] ters challenges, including the development of biocompatible con-
shear stress, droplet impact velocity, droplet volume and poly- ductive inks (demanding meticulous selection and optimization
mer concentration in jetting-based bioprinting[141] and the wave- of conductive biomaterials) and maintaining high cell viability
length and intensity of light, exposure time, type and concen- during printing (requiring careful selection of techniques and
tration of photo-initiators and presence of unreacted free rad- precise calibration of parameters). Effectively addressing these
icals in vat photopolymerization-based bioprinting[142] signifi- challenges involves navigating a complex optimization prob-
cantly affect the cell viability post-printing. Recent studies have lem, balancing factors such as electronic material conductivity,

Adv. Mater. 2024, 2310006 2310006 (45 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 25. An overview of ML application in typical bioprinting processes. ML can be used to a) generate high-resolution images, b) perform image
segmentation, c) control cell quality, d) optimize printing parameters, e) monitor and correct bioprinting process, f) optimize co-culture medium, and
g) optimize external stimuli for tissue conditioning process. Reproduced with permission.[136a] 2020, Taylor & Francis.

Adv. Mater. 2024, 2310006 2310006 (46 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Figure 26. A) Schematic illustration of the material extrusion-based 3D printing that features highly customizable inks based on versatile materials
to construct all main building blocks of the wearable electronic (e3 )-skin with multimodal sensing and power management capabilities. B) Schematic
illustration of material extrusion printing procedures to prepare 2D and 3D architectures. Top right inset, typical rheological properties of printable
inks; bottom, optical images of as printed 2D and 3D MXene architectures. G0 , storage modulus; G00 , loss modulus. Scale bars, 2 mm. C) A machine
learning framework for multimodal e3 -skin. (D and E) Figure showing the 3D-printed electronic skin. F–M) On-body evaluation using e3-skin for real-
time health surveillance and ML-powered health condition prediction. F) Multiplexed multimodal physiological monitoring of a subject after consuming
an alcoholic beverage with different doses. G) Workflow of the cued Go/No-Go task for quantitative measurements of the deviation in reaction time
(RT) and commission errors (%) for the degree of impairment (DI) for inhibitory control due to the influence of alcohol. H–M) The actual performance
versus ML-predicted RT-H (J) and Error-V (L), and the corresponding SHAP decision plot explaining how each regression model arrives at final task
performance outcome of RT-H (K) and Error-V (M). Reproduced with permission.[153] 2023, American Association for the Advancement of Science.

platform printability, biocompatibility, and cell survival rates. A 4.3.3. Construction

recent demonstration showcased the potential application of ML
in bioelectronics printing (Figure 26).[153] ML emerges as a pow- ML finds application in diverse facets of construction printing, in-
erful tool, assisting researchers in identifying patterns among cluding architectural design, structural analysis, structural health
various parameters and guiding decision-making to optimize monitoring, and durability.[155] In the realm of architectural de-
performance in bioelectronics platforms.[154] sign, ML plays a pivotal role in generating both 2D and 3D

Adv. Mater. 2024, 2310006 2310006 (47 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

innovative layouts. ML models excel in classifying architectural influence the release of drugs over time, contributing to more
styles and recognizing building components from drawings. A informed and efficient drug development processes.
recent study showcased the utilization of the House-GAN model, Furthermore, another captivating application is the optimiza-
demonstrating its capacity to explore new designs by learning tion of loading efficiency for 3D-printed drugs through the uti-
from existing data, generating diverse house layouts based on in- lization of ML models and advanced Design of Experiments
put sketches.[156] Additionally, techniques such as semantic seg- (DOE) techniques.[165] By leveraging ML algorithms and sys-
mentation and CNNs are employed to analyze architectural draw- tematic experimentation, researchers can fine-tune the param-
ings, identify space usage, and simulate interior layouts.[157] eters influencing loading efficiency, ensuring that the maximum
In the domain of structural analysis, ML algorithms are in- amount of drug is effectively incorporated into the 3D-printed
strumental in predicting structural behavior (such as seismic re- structure. The marriage of 3D drug printing and ML models
sponse, buckling, and fatigue analysis). Neural networks are em- holds immense promise for the future of pharmaceutical re-
ployed to predict material properties and assess structural com- search and development, paving the way for more personalized
ponents, contributing to the safety and stability evaluation of ex- and efficient drug therapies.
isting structures.[158] Furthermore, ML, coupled with vibration-
based data, facilitates the detection, localization, and quantifica-
tion of damage in steel beams.[159] Notably, ML can also be ap- 4.3.5. Electronics
plied to structural health monitoring for 3D-printed buildings,
evaluating the condition of structures over time, detecting de- 3D-printed electronics have emerged as a ground-breaking fron-
fects, deterioration, and potential failures.[160] ML algorithms an- tier in AM, introducing novel opportunities for integrating cir-
alyzed sensor data from strain gauges and accelerometers to pre- cuits, sensors, and devices directly within printed structures.[166]
dict structural health, identify anomalies, and provide recom- ML plays a dual role in this domain: 1) optimizing the printing
mendations for maintenance or repairs.[161] Durability assess- processes and detecting anomalies during fabrication, 2) process-
ment, considering factors like material properties, environmen- ing data collected from the 3D-printed sensors. These printed
tal exposure, and load-bearing capacity, benefits from ML by pre- sensors, enhanced by ML algorithms, can exhibit adaptive behav-
dicting the lifespan of structures and optimizing their design for iors, dynamically responding to their environment and ensuring
longevity. optimal performance in end-use applications. The synergy be-
Despite notable progress, challenges such as data quality, tween 3D printing and ML thus presents a unique avenue for
model interpretability, and real-world implementation necessi- creating intelligent electronics that are both fabricated and func-
tate careful consideration. The ongoing evolution of ML is poised tionally enhanced by advanced computational techniques.
to have a profound impact on the construction industry, reshap- The advances in additive nanomanufacturing of flexible wear-
ing the entire lifecycle of structures from conceptual design to op- able electronics have been presented, showcasing the potential
erational maintenance. ML-driven innovations are ushering in a of printed bioelectronic systems for portable healthcare, human-
revolution in construction, facilitating the creation of safer, more machine interfaces, and advanced wearable technologies.[167]
efficient, and sustainable built environments. Aerosol jet printing was used to fabricate soft electromyogra-
phy (EMG) electrodes for recording signals from the skin and
CNN was applied for pose-prediction. The results achieved over
4.3.4. Drug 97% accuracy in classifying six muscle activities, enabling real-
time, wireless control of external machines. In another work, a
To date, 3D printing technology has garnered increasing atten- graphene-based electrode was fabricated, and a similar ML tech-
tion within the pharmaceutical sector, revolutionizing drug man- nique was applied for pose prediction (Figure 22). It demon-
ufacturing. One of the key advantages of drug printing lies in strated about 99% accuracy in detecting seven classes of fin-
its ability to facilitate production in small batches, offering un- ger motions, facilitating wireless control of a robotic hand. Both
precedented flexibility in customized dosages, geometries, di- studies emphasize the reliability, mechanical flexibility, and high-
mensions, and controllable drug release profiles. This break- fidelity recording capabilities of the printed bioelectronic sys-
through in manufacturing capability leads to the on-demand fab- tems. The integration of ML algorithms enhances classification
rication of personalized medicines. Remarkably, drug printing accuracy and ensures precise control and continuous monitoring
finds applications across the entire spectrum of the drug devel- in wearable devices. These quantitative findings validate the fea-
opment process, ranging from preclinical drug development and sibility and effectiveness of these technologies in revolutionizing
human clinical trials to the actual intake of medicines.[162] healthcare and human performance.
Numerous studies have reported the use of ML to optimize the A novel wearable biosensing system that used surface EMG
printing parameters in the drug printing process.[163] Notably, an and hyperdimensional computing for real-time hand gesture
intriguing application of ML involves predicting the drug dissolu- recognition was demonstrated.[168] The device comprised of a
tion behavior of 3D-printed medicine based on the drug’s compo- screen-printed, conformal electrode array and custom-designed
sition. Several studies have demonstrated the capability to predict application-specific integrated circuit and it incorporated adap-
drug dissolution profiles by considering various input parame- tive learning and inference capabilities within the sensor. It
ters, including material composition, glass transition tempera- classified 13 hand gestures with 97.12% accuracy and main-
ture, melting temperature, molecular weight, infill pattern, den- tained high accuracy (92.87%) even when expanded to 21 ges-
sity, and surface area-to-volume ratio.[164] This predictive mod- tures. The device facilitated real-time updates of its ML models
eling proves invaluable in understanding how different factors to adapt to changes such as different arm positions or sensor

Adv. Mater. 2024, 2310006 2310006 (48 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

replacement, recovering accuracy by 9.5% without needing addi- 5. Concluding Remarks

tional external computation. The system offers potential advance-
ments in human-machine interface applications, allowing fast The integration of ML in AM processes has attracted increasing
initial training and on-the-fly adaptation, and future work could attention due to its superior performance for various AM-related
consider additional situational contexts and gesture transitions, applications; the ML models can recognize complex patterns
potentially improving classification performance. The low-cost, from large, curated datasets and elucidate the complex relation-
low-complexity design could also be adapted for other physio- ships among different parameters to improve decision-making
logical signal processing applications, like electrocardiography or during the AM process. Some common ML applications in AM
electroencephalography (Figure 27). research include quality control, process optimization, design op-
timization, microstructure analysis, and material formulation.
The implementation of ML in AM helps to enhance the efficiency
4.3.6. Food and reliability of AM processes. Quality control involves the col-
lection of signals from in situ sensors to train ML models for
In recent years, the global food industry has been at the fore- monitoring process stability and detecting defects within printed
front of a transformative paradigm shift, responding to press- layers. Process optimization relies on large datasets from previ-
ing challenges such as environmental sustainability, animal wel- ous printing runs for the prediction of optimal process parame-
fare, and the escalating demand for protein-rich diets. This shift ters under a given set of conditions. The incorporation of ML in
has given rise to the innovative concept of 3D cultivated meat, design optimization enables training on a myriad of design varia-
often referred to as lab-grown or cell-based meat. This cutting- tions, learning from previous iterations, and providing real-time
edge approach to meat production is currently in its infancy, feedback during the design phase. Advanced imaging techniques
with researchers working intensively to surmount technical, cost, were used to collect vast amounts of data which can be processed
and regulatory hurdles. The ultimate goal is to provide a sus- and extracted easily using ML algorithms to decipher the rela-
tainable and ethical alternative to conventional meat produc- tionships between the processing parameters and microstruc-
tion, addressing the growing concerns associated with traditional ture variations. ML algorithms can be used to guide the selection
practices.[169] of optimal material compositions to achieve desired properties
Notably, the incorporation of ML offers a myriad of advan- within the processability constraints. Furthermore, ML applica-
tages to produce 3D-printed cultivated meat. These advantages tions in emerging AM-related fields such as bioprinting, bioelec-
span from the meticulous fabrication of 3D meat-like struc- tronics, construction printing, drug printing, electronics print-
tures to the precise regulation of food texture and the cus- ing, and food printing were highlighted.
tomization of nutritional profiles, ushering in a new era of In AM, even seemingly straightforward calibration can be sub-
precision and customization.[170] Recent studies have already ject to a multitude of variable factors, including environmental
demonstrated the use of ML in various aspects of 3D-printed conditions, equipment wear and tear, and batch-to-batch material
cultivated meat, showing its potential for optimization and en- inconsistencies. ML, particularly adaptive algorithms, can contin-
hancement. These applications range from the optimization of uously learn from new data to adjust for these variations, thereby
culture medium formulation,[171] to the prediction and regula- maintaining and even improving the calibration over time with-
tion of food flavor,[172] and even quality control measures.[173] out manual re-calibration. The AM landscape is rapidly evolving,
A pivotal stage in cultivated meat production lies in cell pro- with the development of new materials and complex geometries
duction within bioreactors, and ML is proving to be a valu- that present high-dimensional challenges suitable for advanced
able tool for real-time adjustments. Parameters such as temper- ML techniques. In these scenarios, ML can be integral not just for
ature, pH, oxygen levels, and nutrient circulation can be dynam- calibration, but for optimizing printing strategies for novel ma-
ically optimized through ML, facilitating optimal cell growth. terials, predicting mechanical properties of printed objects, and
A recent study showcased the application of ML to sustainably enabling real-time quality control for intricate structures that are
optimize serum-free media development, identifying the opti- beyond the capabilities of traditional manufacturing processes.
mal combination of media ingredients that strike a balance be- While current DL research indeed focuses on complex do-
tween yield, environmental impact, and cost for cultivated meat mains such as image and language processing, the principles
production.[171a] and models developed in these domains can be adapted to the
Furthermore, ML also plays a pivotal role in various facets of high-dimensional aspects of AM. For example, CNNs, primarily
3D-cultivated meat production. ML contributes to the develop- used for image data, can be repurposed to analyze topological fea-
ment, optimization, and scale-up of the entire process. By ana- tures of printed layers, and recurrent neural networks can model
lyzing data on different cell types, ML can identify the most suit- time-series data from the printing process to predict and com-
able type of cells for cultured meat production based on factors pensate for potential defects. The intersection of AM with high-
such as growth rate, nutrient requirements, and flavor profile to dimensional data becomes evident in the pursuit of customiz-
achieve the best-quality cultured meat products.[174] In essence, able manufacturing, such as for medical implants or aerospace
ML proves to be an invaluable tool in the development of 3D- components, where bespoke designs are the norms. Here, ML
printed cultivated meat, aiding researchers in overcoming chal- can navigate the vast design space to optimize for specific perfor-
lenges related to cell selection, bioreactor control, product qual- mance criteria, considering individualized constraints and objec-
ity, and nutritional requirements. The role of ML has become in- tives. The application of ML in AM is not limited to simple cal-
creasingly pivotal in shaping the future of sustainable and ethical ibration problems but extends to tackling the intrinsic complex-
meat production. ity and evolving challenges of the field. The potential for ML to

Adv. Mater. 2024, 2310006 2310006 (49 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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

Figure 27. Schematics showing the application of ML to process the output signals from the printed electrodes and sensors. Reproduced with

© 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH

www.advmat.de

2310006 (50 of 56)

permission.[167b] 2020, Nature Portfolio.
www.advancedsciencenews.com

Adv. Mater. 2024, 2310006

 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

contribute to AM is vast and varied, and we anticipate further in- G. Rätsch), Springer, Berlin, Heidelberg 2004; b) T. Hastie, R.
tegration of advanced ML techniques as the technology and ma- Tibshirani, J. Friedman, in The Elements of Statistical Learning: Data
terials of AM continue to advance. Mining, Inference, and Prediction, (Eds: T. Hastie, R. Tibshirani, J.
Lastly, establishing a unified data community could serve as Friedman), Springer, New York 2009.
[4] J. E. Van Engelen, H. H. Hoos, Machine Learning 2020, 109,
a catalyst for overcoming current bottlenecks in ML-driven 3D
373.
printing research. By ensuring that data and insights are freely
[5] a) K. Arulkumaran, M. P. Deisenroth, M. Brundage, A. A. Bharath,
shared, researchers can build on each other’s work, accelerating IEEE Signal Process Mag 2017, 34, 26; b) M. A. Wiering, M. Van
the path from experimental prototypes to practical applications. Otterlo, Adapt., Learn., Optimiz. 2012, 12, 729; c) L. P. Kaelbling,
Additive Manufacturing Materials Database (AMMD) is an ini- M. L. Littman, A. W. Moore, J. Artif. Intell. Res. 1996, 4, 237.
tiative that fosters data sharing and collaboration within the AM [6] G. L. Goh, G. D. Goh, J. W. Pan, P. S. P. Teng, P. W. Kong, Sensors
community.[175] The AMMD enables users to browse and search 2023, 23, 9759.
for specific datasets, facilitating easy access to a wealth of infor- [7] G. L. Goh, X. Huang, W. Toh, Z. Li, S. Lee, W. Y. Yeong, B. S. Han, T.
mation regarding AM materials and processes. For users who Y. Ng, Int. J. AI Mater. Design 2024, 1, 2328.
have data but lack a framework to structure it, AMMD provides a [8] J. Zhang, C. Yin, Y. Xu, S. L. Sing, Int. J. AI Mater. Design 2024, 1,
2301.
data schema and a tool called “Curate Data” to assist in organiz-
[9] L. Scime, J. Beuth, Addit. Manuf. 2018, 19, 114.
ing their information according to standardized practices. The
[10] C. Gobert, E. W. Reutzel, J. Petrich, A. R. Nassar, S. Phoha, Addit.
database is built upon the National Institute of Standards and Manuf. 2018, 21, 517.
Technology’s Material Data Curation System (MDCS), utilizing a [11] C. Liu, R. R. Wang, I. Ho, Z. J. Kong, C. Williams, S. Babu, C. Joslin,
structure defined by NIST’s AM schema. This suggests a robust, J. Intell. Manuf. 2023, 34, 2673.
standards-driven approach to data curation in the field of additive [12] a) P. Yadav, V. K. Singh, T. Joffre, O. Rigo, C. Arvieu, E. Le Guen,
manufacturing. Nonetheless, currently available databases con- E. Lacoste, Adv. Eng. Mater. 2020, 22, 2000660; b) S. Feng, Z.
sist primarily of data extracted from powder bed fusion process Chen, B. Bircher, Z. Ji, L. Nyborg, S. Bigot, Mater. Des. 2022, 222,
and more effort from the AM community to contribute to the 111115.
development of the database. We strongly believe that ML is an [13] H. Baumgartl, J. Tomas, R. Buettner, M. Merkel, Progress Additive
Manuf. 2020, 5, 277.
indispensable tool in realizing the full potential of AM, paving
[14] S. Oster, P. P. Breese, A. Ulbricht, G. Mohr, S. J. Altenburg, J. Intell.
the way for unprecedented innovation and efficiency in AM pro-
Manuf. 2023.
cesses. The integration of ML in AM processes would signifi- [15] S. Oster, T. Fritsch, A. Ulbricht, G. Mohr, G. Bruno, C. Maierhofer,
cantly enhance its efficiency and reliability and amplify the 3D S. J. Altenburg, Metals 2022, 12, 947.
printing capabilities across various AM-related research areas. [16] L. Scime, J. Beuth, Addit. Manuf. 2019, 25, 151.
[17] A. Gaikwad, R. J. Williams, H. de Winton, B. D. Bevans, Z. Smoqi, P.
Rao, P. A. Hooper, Mater. Des. 2022, 221, 110919.
Acknowledgements [18] Z. Ren, L. Gao, S. J. Clark, K. Fezzaa, P. Shevchenko, A. Choi,
W. Everhart, A. D. Rollett, L. Chen, T. Sun, Science 2023, 379,
W.L.Ng would like to acknowledge support from the NTU Presidential 89.
Postdoctoral Fellowship. W.Y.Yeong would like to acknowledge the sup- [19] J. R. Tempelman, A. J. Wachtor, E. B. Flynn, P. J. Depond, J.-B. Forien,
port of the National Research Foundation for NRF Investigatorship Award G. M. Guss, N. P. Calta, M. J. Matthews, Addit. Manuf. 2022, 55,
No.: NRF-NRFI07-2021-0007. Part of this research was supported by 102735.
A*STAR under its RIE2015 JCO Career Development Award (Grant No [20] M. Ghayoomi Mohammadi, D. Mahmoud, M. Elbestawi, Opt. Laser
202D800024).
Technol. 2021, 143, 107338.
[21] S. Luo, X. Ma, J. Xu, M. Li, L. Cao, Sensors 2021, 21, 7179.
[22] R. Drissi-Daoudi, V. Pandiyan, R. Logé, S. Shevchik, G. Masinelli,
Conflict of Interest H. Ghasemi-Tabasi, A. Parrilli, K. Wasmer, Virtual Phys. Prototyping
2022, 17, 181.
The authors declare no conflict of interest. [23] K. Wasmer, T. Le-Quang, B. Meylan, S. A. Shevchik, J. Mater. Eng.
Perform. 2019, 28, 666.
[24] V. Pandiyan, G. Masinelli, N. Claire, T. Le-Quang, M. Hamidi-Nasab,
Keywords C. de Formanoir, R. Esmaeilzadeh, S. Goel, F. Marone, R. Logé, S.
Van Petegem, K. Wasmer, Addit. Manuf. 2022, 58, 103007.
additive manufacturing, bioelectronics, bioprinting, construction, culti- [25] Z. Snow, E. W. Reutzel, J. Petrich, J. Mater. Process. Technol. 2022,
vated meat, drug, machine learning
302, 117476.
[26] E. Westphal, H. Seitz, Addit. Manuf. 2022, 50, 102535.
Received: September 26, 2023 [27] G. A. R. Sampedro, D. J. S. Agron, G. C. Amaizu, D.-S. Kim, J.-M.
Revised: March 1, 2024 Lee, Appl. Sci. 2022, 12, 8753.
Published online: [28] Z. Jin, Z. Zhang, G. X. Gu, Manuf. Lett. 2019, 22, 11.
[29] G. D. Goh, N. M. B. Hamzah, W. Y. Yeong, 3D Print. Addit. Manuf.
2022, 10, 428.
[30] L. Tan, T. Huang, J. Liu, Q. Li, X. Wu, Computers Industrial Eng. 2023,
[1] a) M. D. Xames, F. K. Torsha, F. Sarwar, J Intell Manuf 2022, 1; b) G. 176, 108887.
D. Goh, S. L. Sing, W. Y. Yeong, Artif. Intell. Rev. 2021, 54, 63. [31] D. A. J. Brion, S. W. Pattinson, Nat. Commun. 2022, 13, 4654.
[2] R. Caruana, A. Niculescu-Mizil, ICML ’06 2006, 161. [32] O. Davtalab, A. Kazemian, X. Yuan, B. Khoshnevis, J Intell Manuf
[3] a) Z. Ghahramani, in Advanced Lectures on Machine Learning: 2022, 33, 771.
ML Summer Schools 2003, (Eds: O. Bousquet, U. von Luxburg, [33] S. Garfo, M. Muktadir, S. Yi, J. Mechatron. Robot 2020, 4, 74.

Adv. Mater. 2024, 2310006 2310006 (51 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

[34] J. Villacrés, R. Guamán, O. Menéndez, F. A. Cheein, presented at [67] F. P. Brishty, R. Urner, G. Grau, Flexible Printed Electron. 2022, 7,
IECON 2021 – 47th Annual Conference of the IEEE Industrial Elec- 015009.
tronics Society, Toronto, ON, October 2021. [68] S. Ghosh, M. V. Johnson, R. Neupane, J. Hardin, J. D. Berrigan, S.
[35] D. Ahlers, presented at 2021 International Solid Freeform Fabrication R. Kalidindi, Y. L. Kong, Flexible Printed Electron. 2022, 7, 014011.
Symposium, Austin, TX, August 2021. [69] G. L. Goh, H. Zhang, G. D. Goh, W. Y. Yeong, T. H. Chong, Mater.
[36] A. F. Bonatti, G. Vozzi, C. K. Chua, C. De Maria, Int. J. Bioprint. 2022, Sci. Addit. Manuf. 2022, 1.
8, 620. [70] a) H. Zhang, J. P. Choi, S. K. Moon, T. H. Ngo, Addit. Manuf. 2020, 33,
[37] A. Gaikwad, T. Chang, B. Giera, N. Watkins, S. Mukherjee, A. Pascall, 101096; b) H. Zhang, J. P. Choi, S. K. Moon, T. H. Ngo, J. Mater. Pro-
D. Stobbe, P. Rao, J. Intell. Manuf. 2022, 33, 2093. cess. Technol. 2020, 285, 116779; c) H. Zhang, J. P. Choi, S. K. Moon,
[38] M. Ogunsanya, J. Isichei, S. K. Parupelli, S. Desai, Y. Cai, Proc. Manuf. T. H. Ngo, Advanced Engineering Informatics 2021, 48, 101264; d) H.
2021, 53, 427. Zhang, E. Hong, X. Chen, Z. Liu, ACS Appl. Mater. Interfaces 2023,
[39] S. Stoyanov, C. Bailey, presented at 2017 40th Int. Spring Seminar on 15, 14532.
Electronics Technology (ISSE), Sofia, Bulgaria, May 2017. [71] a) Z. Zhou, H. Shen, B. Liu, W. Du, J. Jin, J. Manuf. Processes 2021,
[40] S. Stoyanov, C. Bailey, G. Tourloukis, presented at 2016 6th Elec- 64, 960; b) K. Ren, Y. Chew, Y. Zhang, J. Fuh, G. Bi, Comput Methods
tronic System-Integration Technology Conference (ESTC), Grenoble, Appl Mech Eng 2020, 362, 112734.
France, September, 2016. [72] J. Guan, S. You, Y. Xiang, J. Schimelman, J. Alido, X. Ma, M. Tang, S.
[41] X. Li, S. Siahpour, J. Lee, Y. Wang, J. Shi, Proc. Manuf. 2020, 48, Chen, Biofabrication 2021, 14, 015011.
643. [73] H. He, Y. Yang, Y. Pan, J Manuf Syst 2019, 50, 236.
[42] Z. Zhang, Z. Liu, D. Wu, Addit. Manuf. 2021, 37, 101692. [74] F. Parisi, V. Sangiorgio, N. Parisi, A. M. Mangini, M. P. Fanti, J. M.
[43] H. Lee, W. Heogh, J. Yang, J. Yoon, J. Park, S. Ji, H. Lee, J Manuf Syst Adam, VINYLTEC 2005, PVC Build. Constr 2023, ahead-of-print.
2022, 62, 575. [75] H. Ko, P. Witherell, Y. Lu, S. Kim, D. W. Rosen, Addit. Manuf. 2021,
[44] J. Ertveldt, P. Guillaume, J. Helsen, Proc. CIRP 2020, 94, 456. 37, 101620.
[45] F. Kaji, H. Nguyen-Huu, A. Budhwani, J. A. Narayanan, M. Zimny, E. [76] T. Xue, T. J. Wallin, Y. Menguc, S. Adriaenssens, M. Chiaramonte,
Toyserkani, J. Manuf. Processes 2022, 81, 624. Extreme Mech Lett 2020, 41, 100992.
[46] L. Chen, X. Yao, P. Xu, S. K. Moon, G. Bi, Virtual Phys, Prototyping [77] S. C.-y. Shen, M. J. Buehler, Communications Engineering 2022, 1, 37.
2021, 16, 50. [78] K. Yanamandra, G. L. Chen, X. Xu, G. Mac, N. Gupta, Compos. Sci.
[47] M. Montazeri, A. R. Nassar, C. B. Stutzman, P. Rao, Addit. Manuf. Technol. 2020, 198, 108318.
2019, 30, 100916. [79] L. Zhou, X. Wu, C. Gu, Journal of Vibroengineering 2015, 17, 3160.
[48] a) L. Chen, X. Yao, S. K. Moon, Mater Today Proc 2022, 70, 136; b) L. [80] N. S. Iyer, A. M. Mirzendehdel, S. Raghavan, Y. Jiao, E. Ulu,
Chen, X. Yao, C. Tan, W. He, J. Su, F. Weng, Y. Chew, N. P. H. Ng, S. M. Behandish, S. Nelaturi, D. M. Robinson, arXiv preprint
K. Moon, Addit. Manuf. 2023, 69, 103547. arXiv:2112.04552 2021.
[49] V. Akhil, G. Raghav, N. Arunachalam, D. Srinivas, J Comput Inf Sci [81] G. D. Goh, S. L. Sing, Y. F. Lim, J. L. J. Thong, Z. K. Peh, S. R. Mogali,
Eng 2020, 20, 021010. W. Y. Yeong, Mater. Des. 2021, 211, 110125.
[50] X. Zhang, J. Saniie, W. Cleary, A. Heifetz, JOM 2020, 72, 4682. [82] Y. Han, R. J. Griffiths, Z. Y. Hang, Y. Zhu, J. Mater. Res. 2020, 35,
[51] D. Cannizzaro, A. G. Varrella, S. Paradiso, R. Sampieri, E. Macii, E. 1936.
Patti, S. Di Cataldo, presented at 2021 Design, Automation & Test in [83] a) C. Herriott, A. D. Spear, Comput. Mater. Sci. 2020, 175, 109599;
Europe Conference & Exhibition (DATE), February 2021. b) J. Li, Z. Yang, G. Qian, F. Berto, Int J Fatigue 2022, 158,
[52] V. Mahato, M. A. Obeidi, D. Brabazon, P. Cunningham, J Intell Manuf 106764.
2022, 33, 845. [84] E. Hu, I. P. Seetoh, C. Q. Lai, Int J Mech Sci 2022, 221, 107190.
[53] Y. Chen, H. Wang, Y. Wu, H. Wang, Materials 2020, 13, 5063. [85] Z. Zhan, H. Li, Int J Fatigue 2021, 142, 105941.
[54] D. Kim, T. Zohdi, Computational Mechanics 2022, 69, 383. [86] S. Liu, A. P. Stebner, B. B. Kappes, X. Zhang, Addit. Manuf. 2021, 39,
[55] P. D. Nguyen, T. Q. Nguyen, Q. B. Tao, F. Vogel, H. Nguyen-Xuan, 101877.
Virtual Phys. Prototyping 2022, 17, 768. [87] A. International, ASTM F2971-13 Standard Practice for Reporting
[56] X. Li, M. Zhang, M. Zhou, J. Wang, W. Zhu, C. Wu, X. Zhang, J. Data for Test Specimens Prepared by Additive Manufacturing, 2021.
Manuf. Processes 2023, 90, 274. [88] A. International, ASTM E8/E8M-22 Standard Test Methods for Ten-
[57] K. Ruberu, M. Senadeera, S. Rana, S. Gupta, J. Chung, Z. Yue, S. sion Testing of Metallic Materials, 2022.
Venkatesh, G. Wallace, Appl. Mater. Today 2021, 22, 100914. [89] A. International, ASTM E9-19 Standard Test Methods of Compres-
[58] K.-T. Fang, D. K. Lin, P. Winker, Y. Zhang, Technometrics 2000, 42, sion Testing of Metallic Materials at Room Temperature, 2019.
237. [90] D. J. Huang, H. Li, Mater. Des. 2021, 203, 109606.
[59] Z. Fu, V. Angeline, W. Sun, Int J Bioprint 2021, 7, 434. [91] J. Zhang, P. Wang, R. X. Gao, Comput Ind 2019, 107, 11.
[60] J. J. Ong, B. M. Castro, S. Gaisford, P. Cabalar, A. W. Basit, G. Pérez, [92] A. K. Gupta, M. Taufik, Austr. J. Mech. Eng. 2022, 1.
A. Goyanes, Int. J. Pharm: X 2022, 4, 100120. [93] R. Cai, H. Lin, P. Cheng, Z. Zhang, K. Wang, Y. Peng, Y. Wu, S. Ahzi,
[61] I. Rojek, D. Mikołajewski, P. Kotlarz, K. Tyburek, J. Kopowski, E. Polym. Compos. 2022, 43, 5235.
Dostatni, Materials 2021, 14, 7625. [94] S. Czarnecki, Ł. Sadowski, J. Hoła, Autom Constr 2021, 132,
[62] T. S. Tamir, G. Xiong, Q. Fang, Y. Yang, Z. Shen, M. Zhou, J. Jiang, 103977.
Int. J. Computer Integr. Manuf. 2022, 1. [95] A. Bagheri, C. Cremona, Mater. Adv. 2020, 1, 720.
[63] W. Lao, M. Li, T. N. Wong, M. J. Tan, T. Tjahjowidodo, Virtual Model. [96] H. Izadgoshasb, A. Kandiri, P. Shakor, V. Laghi, G. Gasparini, Appl.
Rapid Manuf.: Adv. Res. Virtual Rapid Prototyping, Proc. Int. Conf. Adv. Sci. https://doi.org/10.3390/app112210826.
Res. Rapid Prototyping, 2nd 2020, 15, 178. [97] K. Tran-Quoc, L. B. Nguyen, V. H. Luong, H. Nguyen-Xuan, Vietnam
[64] J. Shi, J. Song, B. Song, W. F. Lu, Engineering 2019, 5, 586. J. Mech. 2022, 44, 538.
[65] D. Wu, C. Xu, J. Manuf. Sci. Eng. 2018, 140, 101007. [98] G. A. Lyngdoh, N.-K. Kelter, S. Doner, N. M. A. Krishnan, S. Das,
[66] M. Abbasi Shirsavar, M. Taghavimehr, L. J. Ouedraogo, M. Mater. Des. 2022, 213, 110341.
Javaheripi, N. N. Hashemi, F. Koushanfar, R. Montazami, Biosens. [99] Z. Chang, Z. Wan, Y. Xu, E. Schlangen, B. Šavija, Eng. Fract. Mech.
Bioelectron. 2022, 212, 114418. 2022, 271, 108624.

Adv. Mater. 2024, 2310006 2310006 (52 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

[100] a) C. D. Boley, S. C. Mitchell, A. M. Rubenchik, S. S. Q. Wu, Appl. [132] J. Urlings, G. de Jong, T. Maal, D. Henssen, J Digit Imaging 2023,
Opt. 2016, 55, 6496; b) J. Trapp, A. M. Rubenchik, G. Guss, M. J. 1.
Matthews, Appl. Mater. Today 2017, 9, 341. [133] a) P. Zhuang, W. L. Ng, J. An, C. K. Chua, L. P. Tan, PLoS One 2019,
[101] G. Bi, C. N. Sun, A. Gasser, J. Mater. Process. Technol. 2013, 213, 463. 14, 0216776; b) I. T. Ozbolat, M. Hospodiuk, Biomaterials 2016, 76,
[102] J. H. Martin, B. D. Yahata, J. M. Hundley, J. A. Mayer, T. A. Schaedler, 321; c) W. L. Ng, W. Y. Yeong, M. W. Naing, Proc. of the 1st Int. Conf.
T. M. Pollock, Nature 2017, 549, 365. on Progress in Additive Manufacturing 2014, 441; d) W. L. Ng, W. Y.
[103] P. C. Collins, D. A. Brice, P. Samimi, I. Ghamarian, H. L. Fraser, Annu. Yeong, M. W. Naing, Int J Bioprint 2016, 2, 53; e) W. L. Ng, W. Y.
Rev. Mater. Res. 2016, 46, 63. Yeong, M. W. Naing, Proc. CIRP 2016, 49, 105.
[104] C. A. Brice, W. A. Tayon, J. A. Newman, M. V. Kral, C. Bishop, A. [134] a) R. Suntornnond, W. L. Ng, X. Huang, E. C. H. Yeow, W. Y. Yeong,
Sokolova, Mater. Charact. 2018, 143, 50. J. Mater. Chem. B 2022, 10, 5989; b) W. L. Ng, W. Y. Yeong, M. W.
[105] a) J. A. Slotwinski, E. J. Garboczi, P. E. Stutzman, C. F. Ferraris, S. S. Naing, Proc. of the Int. Conf. on Progress in Additive Manufacturing
Watson, M. A. Peltz, J Res Natl Inst Stand Technol 2014, 119, 460; b) 2016, 397; c) W. L. Ng, J. M. Lee, W. Y. Yeong, M. W Naing, Biomater.
A. Strondl, O. Lyckfeldt, H. Brodin, U. Ackelid, JOM 2015, 67, 549. Sci. 2017, 5, 632.
[106] P. Villars, N. Onodera, S. Iwata, J. Alloys Compd. 1998, 279, 1. [135] W. Li, L. S. Mille, J. A. Robledo, T. Uribe, V. Huerta, Y. S. Zhang, Adv.
[107] S. Curtarolo, W. Setyawan, S. Wang, J. Xue, K. Yang, R. H. Taylor, Healthcare Mater. 2020, 9, 2000156.
L. J. Nelson, G. L. W. Hart, S. Sanvito, M. Buongiorno-Nardelli, N. [136] a) W. L. Ng, A. Chan, Y. S. Ong, C. K. Chua, Virtual Phys. Prototyping
Mingo, O. Levy, Comput. Mater. Sci. 2012, 58, 227. 2020, 15, 340; b) W. L. Ng, C. K. Chua, Y.-F. Shen, Prog. Polym. Sci.
[108] A. Jain, S. Ong, G. Hautier, W. Chen, W. Richards, S. Dacek, S. Cholia, 2019, 97, 101145; c) R. Levato, T. Jungst, R. G. Scheuring, T. Blunk,
D. Gunter, D. Skinner, G. Ceder, 2013, 1, 011002. J. Groll, J. Malda, Adv. Mater. 2020, 32, 1906423; d) J. He, M. Mao,
[109] G. D. Goh, X. Huang, S. Huang, J. L. J. Thong, J. J. Seah, W. Y. Yeong, X. Li, C. K. Chua, Int J Bioprint 2021, 7, 395; e) W. L. Ng, S. Wang, W.
Mater. Sci. Addit. Manuf. 2023, 2. Y. Yeong, M. W. Naing, Trends Biotechnol. 2016, 34, 689; f) J. M. Lee,
[110] M. Charrier, C. M. Ouellet-Plamondon, Cem. Concr. Res. 2022, 156, W. L. Ng, W. Y. Yeong, Appl Phys Rev 2019, 6, 011307; g) W. L. Ng, J.
106761. M. Lee, M. Zhou, W. Y. Yeong, in Rapid Prototyping of Biomaterials,
[111] S. Nazar, J. Yang, M. Faisal Javed, K. Khan, L. Li, Q.-f. Liu, Structures (Ed.: R. Narayan), Elsevier, Chapel Hill, NC 2020.
2023, 48, 1670. [137] A. Tulsyan, G. Schorner, H. Khodabandehlou, T. Wang, M. Coufal,
[112] I. Navarrete, I. L. Fé-Perdomo, J. A. Ramos-Grez, M. Lopez, Constr. C. Undey, Biotechnol. Bioeng. 2019, 116, 2575.
Build. Mater. 2023, 371, 130632. [138] Y. Li, C. M. Nowak, U. Pham, K. Nguyen, L. Bleris, npj Syst. Biol. Appl.
[113] C. Hu, W. N. G. J. Hau, W. Chen, Q.-H. Qin, J. Compos. Mater. 2020, 2021, 7, 23.
55, 1459. [139] a) Y. Lu, R. Rai, N. Nitin, Food Res Int 2023, 173, 113384; b) H. Chen,
[114] D. Cipollone, H. Yang, F. Yang, J. Bright, B. Liu, N. Winch, N. Wu, K. Y. Liu, S. Balabani, R. Hirayama, J. Huang, Research 2023, 6, 0197; c)
A. Sierros, J. Mater. Process. Technol. 2021, 295, 117159. D. Oh, M. Shirzad, M. C. Kim, E.-J. Chung, S. Y. Nam, Int J Bioprint
[115] H. Zhang, Z. Liu, S. Yin, H. Xu, Sci. Rep. 2023, 13, 2513. 2023, 9, 1280.
[116] M. J. Buehler, Appl. Mech. Rev. 2024, 76. [140] a) S. Han, C. M. Kim, S. Jin, T. Y. Kim, Biofabrication 2021, 13, 035048;
[117] K. Margatina, T. Schick, N. Aletras, J. Dwivedi-Yu, presented at Find- b) L. Fischer, M. Nosratlo, K. Hast, E. Karakaya, N. Ströhlein, T. U.
ings of the Association for Computational Linguistics: EMNLP 2023, Esser, R. Gerum, S. Richter, F. Engel, R. Detsch, Biofabrication 2022,
Singapore, 2023. 14, 045005.
[118] S. Badini, S. Regondi, E. Frontoni, R. Pugliese, Adv. Ind. Eng. Polym. [141] a) W. L. Ng, V. Shkolnikov, Int J Bioprint 2024, 2135; b) A. Blaeser,
Res. 2023, 6, 278. D. F. Duarte Campos, U. Puster, W. Richtering, M. M. Stevens, H.
[119] Y.-C. Hsu, Z. Yang, M. J. Buehler, APL Mater. 2022, 10. Fischer, Adv. Healthcare Mater. 2016, 5, 326; c) W. L. Ng, X. Huang,
[120] Y. Zhang, G. S. Hong, D. Ye, K. Zhu, J. Y. Fuh, Mater. Des. 2018, 156, V. Shkolnikov, G. L. Goh, R. Suntornnond, W. Y. Yeong, Int J Bioprint
458. 2022, 8, 424; d) W. L. Ng, X. Huang, V. Shkolnikov, R. Suntornnond,
[121] Y. Zhang, H. G. Soon, D. Ye, J. Y. H. Fuh, K. Zhu, IEEE Trans Industr W. Y. Yeong, Bio-Des. Manuf 2023, 6, 676; e) W. L. Ng, W. Y. Yeong,
Inform 2019, 16, 5769. M. W. Naing, Materials 2017, 10, 190.
[122] S. A. Shevchik, C. Kenel, C. Leinenbach, K. Wasmer, Addit. Manuf. [142] W. L. Ng, J. M. Lee, M. Zhou, Y.-W. Chen, K.-X. A. Lee, W. Y. Yeong,
2018, 21, 598. Y.-F. Shen, Biofabrication 2020, 12, 022001.
[123] M. Montazeri, A. R. Nassar, A. J. Dunbar, P. Rao, IISE Trans 2020, [143] a) D. Mohammadrezaei, L. Podina, J. De Silva, M. Kohandel, Bio-
52, 500. fabrication 2023; b) H. Xu, Q. Liu, J. Casillas, M. McAnally, N.
[124] W. Ren, G. Wen, Z. Zhang, J. Mazumder, Mater. Manuf. Processes Mubtasim, L. S. Gollahon, D. Wu, C. Xu, J Intell Manuf 2022, 33,
2022, 37, 1339. 995.
[125] M. Khanzadeh, S. Chowdhury, M. Marufuzzaman, M. A. Tschopp, [144] X. Huang, W. L. Ng, W. Y. Yeong, J Intell Manuf 2023.
L. Bian, J Manuf Syst 2018, 47, 69. [145] a) B.-S. Kim, J. Nikolovski, J. Bonadio, D. J. Mooney, Nat. Biotechnol.
[126] J. S. Kim, C. S. Lee, S.-M. Kim, S. W. Lee, Int. J. Precision Eng. Manuf. 1999, 17, 979; b) D. Seliktar, R. A. Black, R. P. Vito, R. M. Nerem, Ann.
Green Technol. 2018, 5, 479. Biomed. Eng. 2000, 28, 351.
[127] Z. Ye, C. Liu, C. Kan, Manuf. Lett. 2023, 35, 1155. [146] a) A. Llucià-Valldeperas, C. Soler-Botija, C. Gálvez-Montón, S.
[128] a) T. DebRoy, T. Mukherjee, H. Wei, J. Elmer, J. Milewski, Nat. Rev. Roura, C. Prat-Vidal, I. Perea-Gil, B. Sanchez, R. Bragos, G. Vunjak-
Mater. 2021, 6, 48; b) S. Nasiri, M. R. Khosravani, J Mater Res Technol Novakovic, A. Bayes-Genis, Stem Cells Transl. Med. 2017, 6, 970; b)
2021, 14, 1137. K. Ronaldson-Bouchard, S. P. Ma, K. Yeager, T. Chen, L. Song, D.
[129] F. García-Martínez, D. Carou, F. de Arriba-Pérez, S. García-Méndez, Sirabella, K. Morikawa, D. Teles, M. Yazawa, G. Vunjak-Novakovic,
Rapid Prototyping J. 2023. Nature 2018, 556, 239.
[130] O. Ero, K. Taherkhani, E. Toyserkani, Addit. Manuf. 2023, 78, 103894. [147] a) A. Satyam, P. Kumar, X. Fan, A. Gorelov, Y. Rochev, L. Joshi, H.
[131] a) N. Rane, S. Choudhary, J. Rane, Virtual Reality (VR), Augmented Peinado, D. Lyden, B. Thomas, B. Rodriguez, Adv. Mater. 2014, 26,
Reality (AR) D 2023, 4; b) L. O. S. Bozzi, K. D. Samson, S. Tadeja, 3024; b) W. L. Ng, M. H. Goh, W. Y. Yeong, M. W. Naing, Biomater.
T. Bohné, S. Pattinson, presented at 2023 IEEE Conference on Virtual Sci. 2018, 6, 562; c) P. Benny, C. Badowski, E. B. Lane, M. Raghunath,
Reality and 3D User Interfaces Abstracts and Workshops (VRW), 2023. Tissue Eng Part A 2014, 21, 183.

Adv. Mater. 2024, 2310006 2310006 (53 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

[148] a) M. Pruniéras, M. Régnier, D. Woodley, J. Invest. Dermatol. 1983, [163] a) M. Elbadawi, B. M. Castro, F. K. Gavins, J. J. Ong, S. Gaisford,
81, S28; b) W. L. Ng, Z. Q. Tan, W. Y. Yeong, M. W. Naing, Bio- G. Pérez, A. W. Basit, P. Cabalar, A. Goyanes, Int. J. Pharm. 2020,
fabrication 2018, 10, 025005; c) J.-E. Chang, S. K. Basu, V. H. Lee, 590, 119837; b) M. Elbadawi, T. Gustaffson, S. Gaisford, A. W. Basit,
Pharm. Res. 2000, 17, 670; d) S. Wang, C. E. Ghezzi, R. Gomes, R. E. Int. J. Pharm. 2020, 590, 119868; c) Y. Abdalla, M. Elbadawi, M.
Pollard, J. L. Funderburgh, D. L. Kaplan, Biomaterials 2017, 112, 1; Ji, M. Alkahtani, A. Awad, M. Orlu, S. Gaisford, A. W. Basit, Int. J.
e) G. Lacroix, W. Koch, D. Ritter, A. C. Gutleb, S. T. Larsen, T. Loret, Pharm. 2023, 633, 122628; d) C. S. O’Reilly, M. Elbadawi, N. Desai,
F. Zanetti, S. Constant, S. Chortarea, B. Rothen-Rutishauser, Appl. S. Gaisford, A. W. Basit, M. Orlu, Pharmaceutics 2021, 13, 2187; e) P.
In Vitro Toxicol. 2018, 4, 91; f) W. L. Ng, T. C. Ayi, Y.-C. Liu, S. L. Sing, Carou-Senra, J. J. Ong, B. M. Castro, I. Seoane-Viano, L. Rodríguez-
W. Y. Yeong, B.-H. Tan, Int J Bioprint 2021, 7, 332. Pombo, P. Cabalar, C. Alvarez-Lorenzo, A. W. Basit, G. Pérez, A.
[149] a) S. Rey, G. L. Semenza, Cardiovasc. Res. 2010, 86, 236; b) M. A. Goyanes, Int. J. Pharm. X 2023, 5, 100181.
Kuss, R. Harms, S. Wu, Y. Wang, J. B. Untrauer, M. A. Carlson, B. [164] a) T. Tagami, C. Morimura, T. Ozeki, Int. J. Pharm. 2021, 604, 120721;
Duan, RSC Adv. 2017, 7, 29312. b) S. Obeid, M. Madžarević, M. Krkobabić, S. Ibrić, Int. J. Pharm.
[150] a) H. Sasaki, I. Takeuchi, M. Okada, R. Sawada, K. Kanie, Y. Kiyota, 2021, 601, 120507; c) B. M. Castro, M. Elbadawi, J. J. Ong, T.
H. Honda, R. Kato, PLoS One 2014, 9, e93952; b) F. Matsuoka, I. Pollard, Z. Song, S. Gaisford, G. Pérez, A. W. Basit, P. Cabalar, A.
Takeuchi, H. Agata, H. Kagami, H. Shiono, Y. Kiyota, H. Honda, R. Goyanes, J. Controlled Release 2021, 337, 530; d) A. Lu, J. Zhang, J.
Kato, PLoS One 2013, 8, e55082. Jiang, Y. Zhang, B. R. Giri, V. R. Kulkarni, N. H. Aghda, J. Wang, M.
[151] M. Dattatrey, A. K. Tiwari, B. Ghoshal, J. Singh, IEEE Access 2019, 7, Maniruzzaman, Pharm. Res. 2022, 39, 2905.
122561. [165] J. Wang, N. H. Aghda, J. Jiang, A. M. Habib, D. Ouyang, M.
[152] S. Agarwala, J. M. Lee, W. L. Ng, M. Layani, W. Y. Yeong, S. Magdassi, Maniruzzaman, Int. J. Pharm. 2022, 628, 122302.
Biosens. Bioelectron. 2018, 102, 365. [166] a) A. K. Sinha, G. L. Goh, W. Y. Yeong, Y. Cai, Adv. Mater. Interfaces
[153] Y. Song, R. Y. Tay, J. Li, C. Xu, J. Min, E. Shirzaei Sani, G. Kim, W. 2022, 9, 2200621; b) G. L. Goh, V. Dikshit, R. Koneru, Z. K. Peh, W.
Heng, I. Kim, W. Gao, Sci. Adv. 2023, 9, eadi6492. Lu, G. D. Goh, W. Y. Yeong, Int J Adv Manuf Technol 2022, 120, 2573;
[154] G. D. Goh, J. M. Lee, G. L. Goh, X. Huang, S. Lee, W. Y. Yeong, Tissue c) S. Agarwala, G. L. Goh, W. Y. Yeong, 2018; d) G. L. Goh, H. Zhang,
Eng Part A 2023, 29, 20. T. H. Chong, W. Y. Yeong, Adv. Electron. Mater. n/a, 2100445.
[155] S. K. Baduge, S. Thilakarathna, J. S. Perera, M. Arashpour, P. Sharafi, [167] a) Y.-T. Kwon, H. Kim, M. Mahmood, Y.-S. Kim, C. Demolder, W.-H.
B. Teodosio, A. Shringi, P. Mendis, Autom Constr 2022, 141, 104440. Yeo, ACS Appl. Mater. Interfaces 2020, 12, 49398; b) Y.-T. Kwon, Y.-S.
[156] N. Nauata, K.-H. Chang, C.-Y. Cheng, G. Mori, Y. Furukawa, presented Kim, S. Kwon, M. Mahmood, H.-R. Lim, S.-W. Park, S.-O. Kang, J. J.
at Computer Vision–ECCV 2020: 16th European Conference, Glasgow, Choi, R. Herbert, Y. C. Jang, Y.-H. Choa, W.-H. Yeo, Nat. Commun.
UK, August 23–28, 2020, Proceedings, Part I 16, 2020. 2020, 11, 3450.
[157] J. Seo, H. Park, S. Choo, Appl. Sci. 2020, 10, 7347. [168] A. Moin, A. Zhou, A. Rahimi, A. Menon, S. Benatti, G. Alexandrov,
[158] a) M. Jimenez-Martinez, M. Alfaro-Ponce, Int J Fatigue 2019, 124, S. Tamakloe, J. Ting, N. Yamamoto, Y. Khan, Nat. Electron. 2021, 4,
42; b) P. Mandal, Eng. Struct. 2017, 152, 843; c) U. K. Mallela, A. 54.
Upadhyay, Thin-Walled Structures 2016, 102, 158; d) S. Tohidi, Y. [169] M. J. Post, S. Levenberg, D. L. Kaplan, N. Genovese, J. Fu, C. J.
Sharifi, Thin-Walled Structures 2015, 94, 359. Bryant, N. Negowetti, K. Verzijden, P. Moutsatsou, Nat Food 2020,
[159] a) K. H. Padil, N. Bakhary, H. Hao, Mech Syst Signal Process 2017, 83, 1, 403.
194; b) O. Avci, O. Abdeljaber, S. Kiranyaz, M. Hussein, M. Gabbouj, [170] W. L. Ng, J. S. Tan, Int. J. AI Mater. Design 2024, 1, 2279.
D. J. Inman, Mech Syst Signal Process 2021, 147, 107077. [171] a) A. Nikkhah, A. Rohani, M. Zarei, A. Kulkarni, F. A. Batarseh,
[160] a) O. Abdeljaber, O. Avci, M. S. Kiranyaz, B. Boashash, H. Sodano, N. T. Blackstone, R. Ovissipour, Sci. Total Environ. 2023, 164988;
D. J. Inman, Neurocomputing 2018, 275, 1308; b) S. Dorafshan, R. J. b) Z. Cosenza, D. E. Block, K. Baar, Biotechnol J 2021, 16,
Thomas, M. Maguire, Constr. Build. Mater. 2018, 186, 1031; c) W. R. 2100228.
L. d. Silva, D. S. d. Lucena, presented at Proceedings, 2018; d) Y. Bao, [172] H. Ji, D. Pu, W. Yan, Q. Zhang, M. Zuo, Z. Yuyu, Trends Food Sci.
Z. Tang, H. Li, Y. Zhang, Struct Health Monit 2019, 18, 401; e) Y. J. Technol. 2023, 138, 738.
Cha, W. Choi, O. Büyüköztürk, Computer-Aided Civil and Infrastruc- [173] A. Soni, Y. Dixit, M. M. Reis, G. Brightwell, Compr Rev Food Sci Food
ture Engineering 2017, 32, 361. Saf 2022, 21, 3717.
[161] a) C.-c. Liu, J. Liu, L.-j. Liu, presented at 2009 International Confer- [174] J. Wang, S. Ding, C. Da, C. Chen, Z. Wu, C. Li, G. Zhou, C. Tang, J.
ence on Information Engineering and Computer Science, 2009; b) M. Agric. Food Chem. 2023, 71, 18613.
H. Rafiei, H. Adeli, Eng. Struct. 2018, 156, 598. [175] NIST, National Institute of Standards and Technology, Vol. 2024,
[162] J. Vamathevan, D. Clark, P. Czodrowski, I. Dunham, E. Ferran, G. 2024.
Lee, B. Li, A. Madabhushi, P. Shah, M. Spitzer, Nat. Rev. Drug Dis- [176] S. L. Sing, Y. Y. Wai, E. W. Florencia, J. of Alloys and Comp. 2016, 660,
covery 2019, 18, 463. 461.

Adv. Mater. 2024, 2310006 2310006 (54 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Wei Long Ng is a NTU Presidential Postdoctoral Fellow at the Singapore Centre for 3D Printing
(SC3DP), Nanyang Technological University (NTU), Singapore. He received his Ph.D. degree from
School of Aerospace & Mechanical Engineering, Nanyang Technological University under the A*STAR
Graduate Scholarship. Dr. Ng served as a Research Assistant Professor at the HP-NTU Digital Manu-
facturing Corporate Lab, Singapore from 2020 to 2022. His expertise lies in the intersection of additive
manufacturing and machine learning, where he fabricates 3D biomimetic constructs using novel bio-
printing strategies for applications ranging from tissue engineering to the emerging field of cultivated
meat production.

Guo Liang Goh is a Research Fellow at the School of Mechanical and Aerospace Engineering, Nanyang
Technological University, Singapore. He obtained his Ph.D. degree in Mechanical Engineering from
the Nanyang Technological University. He specializes in the innovative fields of 3D printing, additive
manufacturing, 3D-printed electronics and machine learning. His expertise spans wearables, soft
robotics, bioelectronics and more. He has led numerous research projects, secured substantial fund-
ing, and contributed to pioneering advancements in sensor technology, UAVs, and sports technology.

Guo Dong Goh is a scientist at Singapore Institute of Manufacturing Technology (SIMTech). He ob-
tained his Ph.D. degree in Mechanical Engineering from the Nanyang Technological University. His
research interests include using additive manufacturing process to create multifunctional polymer
parts and applying machine learning techniques to advance the field of 3D printing in terms of design
and quality. He has led numerous research projects, secured substantial funding, and contributed to
pioneering advancements in fracture mechanics in 3D printing composites, UAVs, and sports tech-
nology.

Jyi Sheuan Jason Ten is currently the Deputy Group Manager of the Additive Technology Innovation
Group within the Additive Manufacturing Division at the Singapore Institute of Manufacturing Tech-
nology. He received a Ph.D. degree in lasers for manufacturing from the University of Cambridge in
2018. Dr. Ten’s research interests are in additive manufacturing: including in-process monitoring for
defect detection and quality assurance, sustainability of powder within re-use cycles with correlation
to part properties, fine feature part production, multi-material laser powder bed fusion, materials for
additive manufacturing, and machine learning in AM.

Adv. Mater. 2024, 2310006 2310006 (55 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH
 15214095, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/adma.202310006 by North Eastern Regional Institution Of, Wiley Online Library on [03/04/2024]. 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.advmat.de

Wai Yee Yeong is a Professor and Chair of School of Mechanical & Aerospace Engineering, Nanyang
Technological University (NTU). She joined NTU since 2013 and was promoted to Full Professor in
2022. She is currently serving as a Program Director at HP-NTU Digital Manufacturing Corporate Lab,
Editor-in-Chief of International Journal of AI for Materials and Design (IJAMD) and Associate Editor
for Virtual and Physical Prototyping and International Journal of Bioprinting. Her current research
interest focuses on the convergence of 3D printing and machine learning, particularly in the realms of
printing multi-functional structures, printed electronics, and bio-electronic platforms.

Adv. Mater. 2024, 2310006 2310006 (56 of 56) © 2024 The Authors. Advanced Materials published by Wiley-VCH GmbH