







































S. Choppala et al. /Future Technology                                                                                            May 2023| Volume 02 | Issue 02 | Pages 
36-42 

36 

 

 

 

Article 

Next generation of advanced ceramic 3D printers 
Sam Choppala*, Armin Allam, Zichen Fang, Amir Armani 

San Jose State University, California, United States of America 

A R T I C L E   I N F O 
 

Article history: 
Received 08 November 2022  
Received in revised form 
09 December 2022 
Accepted 13 December 2022 
 
Keywords:  
Additive manufacturing, Technical ceramics 
3D Printing, Extrusion, Machine learning 
 
*Corresponding author 
Email address:  sam.choppala@sjsu.edu 
 
 
DOI: 10.55670/fpll.futech.2.2.5 
 

A B S T R A C T 
 

Ceramic on-demand extrusion (CODE) is a novel slurry-based additive 
manufacturing (AM) process for technical ceramics. Extensive characterization 
studies have shown that this process produces dense ceramic specimens with 
relatively improved mechanical properties such as flexural strength, fracture 
toughness, hardness, etc. The objective of the current study was to develop the 
next generation of CODE. The CODE printer created consists of an aluminum 
extrusion frame, a three-axis gantry system, an extruder, and a heat lamp. The 
ceramic slurry is fed to an extruder that prints parts onto a bedplate. The green 
body parts are then subject to postprocessing, including drying, debinding, and 
sintering. Ceramic composites and functionally graded materials are created 
using CODE to further study the process. Furthermore, a real-time deep 
learning defect detection protocol to identify common defects of CODE while 
printing, as well as a control feedback system to implement corrective action 
based on the defect detected, is being developed. 
 

 

1. Introduction 

Technical ceramics are versatile materials used in 
various industries and applications due to their hardness, 
stability at high temperatures, chemical resistance, electrical 
insulation, and more. Examples of common technical ceramic 
materials include various types of oxides, carbides, nitrides, 
and borides. A few examples of present-day applications 
include zirconia-based dental abutments and implants [1], 
calcium phosphate-based synthetic bone grafts [2], and 
silicate-based high-frequency dielectrics for high-bandwidth 
wireless communications [3]. As the use of technical ceramics 
is prevalent in today’s applications, the quality, efficiency, and 
speed of manufacturing these components are crucial. Under 
ceramic manufacturing, there are two main distinct types of 
processes: conventional fabrication and additive 
manufacturing. Another subcategory within conventional 
fabrication includes pressure-less sintering methods and 
high-pressure sintering methods. This subcategory 
distinguishes the conventional fabrication methods by the 
application of external pressure while sintering to densify the 
component further. Examples of conventional fabrication 
methods include gel casting [4], direct foaming [5], isostatic 
pressing [6], slip casting [7], etc. AM methods could perform 
better in creating ceramic components with complex 
geometries and designs [8]. Examples of AM processes 
include stereolithography [9], direct ink writing [10], binder 
jetting [11], and selective laser sintering [12]. These 
processes can be classified based on the type of feedstock 
used, which can be either powder-based or slurry-based. 
Further detailed information regarding ceramic AM 
techniques can be found in [13-15]. Ceramic on-demand 

extrusion (CODE) is a novel slurry-based 3D printing 
technology used to create ceramic components. The main 
procedure includes a green body that is printed in a layer-
wise fashion [16]. After each layer is printed, a heat lamp is 
used to dry the printed layer partially and uniformly. Then, 
layers below the last layer will submerge into an oil bath to 
prevent evaporation from the sides of the part and preserve 
moisture. This process will occur sequentially until the green 
body is completely printed. This paper outlines and discusses 
the CODE, the current progress of the process, as well as the 
artificial intelligence implementation procedure to improve 
ceramic 3D printing with CODE.  

2. Printer Design 

2.1 Mechanical Design 

The main structure of the CODE 3D printer includes 
aluminum extrusions, rails (X, Y, and Z), an extruder, an oil 
bath, a heat lamp, servo motor drivers, and a printing build 
plate displayed in Figure 1. The frame was built by twelve 80 
mm × 80 mm aluminum T-slot extrusions (40-8080, 80/20 
LLC, Columbia City, IN) and four 40 mm × 80 mm aluminum 
T-slot extrusions. The extrusions are connected by twelve L-
brackets and several T-nuts. The 4080 extrusions and L-
brackets enhance the strength and stability of the 3D printer. 
There are three rails with actuators used in this printer. The 
X and Y rails utilize the same 100 W servo motors (MINAS A6 
100W Servo Motor, Panasonic, Osaka, Japan); the Z rail also 
has a 100W servo motor but with an independent brake 
attached. To make the Z rail fully functional, it is necessary to 
have an external power supply to power the brake so that it 
can be released. 

 

 

Future Technology 

Open Access Journal 

https://doi.org/10.55670/fpll.futech.2.2.5 

 

 

 

 

 

 

May 2023| Volume 02 | Issue 02 | Pages 36-42 

Journal homepage: https://fupubco.com/futech 

 

ISSN 2832-0379 

mailto:%20sam.choppala@sjsu.edu
https://doi.org/10.55670/fpll.futech.2.2.5
https://fupubco.com/futech
https://fupubco.com/


S. Choppala et al. /Future Technology                                                                                            May 2023| Volume 02 | Issue 02 | Pages 36-42 

37 

 

 

 

 

Three Panasonic drivers were purchased along with X, Y, 
and Z rails to control the motors in the rails. As displayed in 
Figure 2, the heat lamp is mounted onto the frame, utilizing a 
custom 3D-printed mounting adapter. After the printer prints 
each layer, the heat lamp will turn on automatically, using the 
G code to dry the printed layer. Once a layer is printed, the 
extruder returns to the home position, triggering the heat 
lamp to turn on and partially dry the printed layer. After each 
layer is printed and dried, the printing bed attached to the Z 
(i.e., vertical) rail will move down to merge the newly finished 
layer into the oil tank. 

2.2 Electrical Design 

The code printer user can insert G code into the 
LinuxCNC software to generate motion signals. Linux CNC is 
an open-source CNC software that is commonly used to 
operate milling machines, laser cutters, lathes, and more. 
These signals are output to the MESA 7i76 (MESA Electronics, 
7i76, El Sobrante, CA) daughter board from the MESA 6i25 
(MESA Electronics, 6i25, El Sobrante, CA) board in the 
motherboard of the host computer. To operate the Panasonic 
motors (Panasonic, Minas A6, Osaka, Japan) that drive the 
linear rails in the X, Y, and Z directions, they must be paired 
with MINAS A6 Panasonic drivers. The MESA 7i76 output 
sends step and direction signals to the drivers through the X4 
I/O port on the Panasonic drivers (Panasonic, Minas A6 
Osaka, Japan). 5VDC power is administered to the MESA 7i76 
board from the computer power supply. The MESA 7i76 
board receives power from two external sources. One from a 
pin out of the IEEE 1284 36-pin male (i.e., DB25) cable 
connected to the MESA 7i76 board coming from the MESA 
6i25 board and another 12VDC from an external power 
supply to power the board and field power outputs, 
respectively. The drivers are powered by external AC power, 
and feed power to the servo motors. A flowchart diagram 
displaying the power and communication lines between the 
electrical components of this system is shown in Figure 3. 

 

 

 

 

 

 

Figure 2. Prototype CODE Printer 

 
2.3  Hardware-Software Implementation 

To control the physical hardware system, the LinuxCNC 
operating system is used, along with two different MESA 
boards and a microstep driver. The two MESA boards being 
used are MESA 6i25 and MESA 7i76. The MESA 6i25 board is 
a general-purpose programmable I/O card for the PCIe bus of 
a computer. This board is directly connected to the host 
computer’s motherboard using the PCIe x1 port.  

Figure 1 . Annotated 3D model of a prototype design for the CODE method 



S. Choppala et al. /Future Technology                                                                                            May 2023| Volume 02 | Issue 02 | Pages 36-42 

38 

 

 

 

Furthermore, the MESA 7i76 is classified as a daughter 
board, meaning it connects to and communicates with the 
MESA 6i25, instead of the host computer directly. The MESA 
7i76 connects to the MESA 6i25 using a DB25 cable, as 
discussed in Section 2.2. The MESA 7i76 board is used for 
interfacing with the drivers of the motor directly using step 
and direction interfaces. It sends digital step pulses to control 
the motor drivers. A microstep driver is used to control the 
extruder, as it is a stepper motor. The hardware can be 
controlled by G code thru the LinuxCNC software interface. 
The top row of Figure 3 displays the signal flow from the 
LinuxCNC software to the MESA 7i76, where the drivers 
obtain the final communication signal to operate the motors. 
The main method of controlling the motors with the MESA 
boards is through LinuxCNC being installed onto the host 
computer. The host computer is selected such that the 
desktop latency, when measured with LinuxCNC is as low as 
possible. The host computer has a desktop latency of around 
9000 ns. Within LinuxCNC, the PNC configuration wizard is 
used to set up the connections between the motor drivers and 
the MESA boards, as well as the models of the boards being 
used and their firmware. The setup of the PNC configuration 
results in two files with the following extensions: .hal and .ini. 
These files can be used to set up PID gains, servo period, and 
other variables. The LinuxCNC interface relies on the 
information from the .hal and .ini files to communicate with 
the operating system and the MESA boards. 

3. Preprocessing 

CODE is a slurry-based extrusion process, and the 
feedstock is in the form of a viscous paste. The four main 
materials used in this slurry are as follows: deionized water, 
ceramic powder, dispersant, and binder. Within the slurry, 
water and ceramic powder cover the majority of the final 
volume. Additional modifications can be made to the paste by 
adding or removing certain materials for different printing 
applications.  

 

 
 
 
The amount and ratio of each material in the slurry are 

dependent on the solids loading and the paste viscosity 
desired. The dispersant in the mixture is used to create a 
homogeneous slurry. The purpose of this is such that the 
ceramic particles do not settle at the bottom over time. The 
use of a dispersant allows for more homogenous green bodies 
when printed. The binder is used to both thicken the paste 
and act as a bonding agent within the green bodies when 
printed. The amount of binder is the least with respect to 
volume out of all the materials used within the slurry. The 
preprocessing phase for any CODE printing material would 
involve the same major stages. The general outline is shown 
below. This outline is visually displayed in Figure 4. 
• Step 1. Combine water, dispersant, and ceramic powder. 
• Step 2. Mix in a ball mill with appropriate ceramic milling 

media for uniform paste in a closed container. The time 
depends on the volume of paste, solids loading, size of ball 
mill ceramic beads, and ceramic powder particle size and 
distribution or until the paste is homogenous.  

• Step 3. Add binder after ball milling the mixture. 
• Step 4. Use a vacuum whip mixer to mix the binder into 

the paste uniformly. The time depends on the volume of 
paste and amount of binder or until the paste is 
homogeneous. 

• Step 5. Use a vibratory table to set the paste (i.e., eliminate 
remaining air bubbles) 

4. Processing 

After creating the feedstock, the material is transferred 
into a hopper system that feeds the feedstock into the 
extruder. The extruder is then moved in the X and Y directions 
to complete the first layer of the print using Luer-lock tips for 
precise control of the slurry. After the first layer is completed, 
the heat lamp is turned on until the printed layer is partially 
dried. This stage is followed by the Z-axis moving down such 
that the partially dried layer is submerged in an oil bath. The 
printer then repeats these stages and prints in a layer-wise 
fashion while each layer is appropriately partially dried and 

Figure 3 . CODE printer process diagram 



S. Choppala et al. /Future Technology                                                                                            May 2023| Volume 02 | Issue 02 | Pages 36-42 

39 

 

submerged in an oil bath. After all the layers are printed, the 
printed specimen will be submerged in the same oil bath for 
a certain time duration such that significant gradients in 
mechanical properties and specimen surface qualities are 
avoided. The G code used for this printer follows the standard, 
which is supported by LinuxCNC. The heat lamp is treated the 
same as the coolant such that it is controlled by LinuxCNC 
using the same M code as the coolant function within 
LinuxCNC, as the coolant is not needed for this printer. 
Furthermore, the extruder is treated as a fourth axis such that 
the dosing and extrusion speed is properly controlled with 
the G code. Processing parameters for printing specimens 
include extrusion rate, nozzle diameter, extruder movement 
speed, layer thickness, line spacing, heat lamp height, and 
lamp timing. These variables are all interconnected and are 
determined empirically for various printing applications. 

5. Postprocessing 

The postprocessing phase involves three main stages. 
These stages are visually displayed as a flowchart in Figure 5. 
• Step 1. Send printed green bodies to a humidity chamber 

to dry the green bodies of water. The time depends on the 
solids loading of the paste and the size of the printed 
specimens, and typically takes less than a day. 

• Step 2. Use a furnace to debind the dried specimens. The 
time depends on the amount of binder used and the size 
of the printed specimens, and typically takes less than two 
hours. 

• Step 3. Use a sintering furnace to densify the parts. The 
time and heating rate depend on the material 
composition, particle size, sintering aids, and green body 
density. It typically takes a few hours. 

The debinding stage (Step 2) and the sintering stage 
(Step 3) both use the sintering furnace; however, the sintering 
schedule (i.e., sintering time, sintering temperature, and 
heating rate) are different. The debinding stage will be done 
at a much lower temperature and for a shorter time than the 
sintering stage, as the binder will burn out in a shorter time 
than the full densification of the ceramic specimen. 
Ghazanfari et al. [16,18,19] delve into further specific 
examples of materials, pre-processing, and postprocessing 
parameters.  

 

 

 

 

6. Sample Parts 

Some sample parts printed using CODE are shown in 
Figure 6. Samples (a), (b), (c) in figure 6 are comprised of 
zirconia powder (TZ-3Y-E, Tosoh USA, Inc., Grove City, OH, 
USA) as the main ceramic material, Dolapix (Dolapix CE 64, 
Zschimmer & Schwarz GmbH, Lahnstein, Germany) as a 
dispersant, ammonium hydroxide solution (221228, Sigma 
Aldrich, St. Louis, MO, USA) for pH adjustment, and deionized 
water. Sample (d) in figure 6 is comprised of alumina powder 
(A-16SG; Almatis, Leetsdale, PA) as the main ceramic 
material, ammonium polymethacrylate (DARVAN® C-N; 
Vanderbilt Minerals, Norwalk, CT) as a dispersant, cold-
water-dispersible methylcellulose (Methocel J5M S; Dow 
Chemical Company, Midland, MI) as a binder, and deionized 
water. A similar pre-processing procedure, as described in 
Section 3, was used to prepare the final slurry with 60 vol% 
solids loading for the alumina samples and 50 vol% solids 
loading for the zirconia samples. Figure 6 below displays the 
sample test specimens with varying geometric complexities 
produced with this slurry. The printing parameters used to 
create the zirconia samples used a nozzle diameter of 600 µm, 
300µm, and 200µm, respectively, and the alumina specimen 
used a 610µm nozzle diameter. Line spacing and layer 
thickness parameters were selected based on a trade-off 
between efficiency and accuracy for different parts. Radiation 
distance and heat lamp time have been decided through 
experimentation; however, these variables are constant for 
any size of the layer, as only the top surface of the layer is 
exposed to the heat lamp. 

7. Closed-loop Control Using Deep Learning 

De La Rosa [17] used a convolutional neural network 
(CNN) to detect common printing defects and a closed-loop 
feedback system such that the defects detected would alter 
the print settings for fused filament fabrication (FFF). 
Equation 1 describes the process of convolution that the 
neural network implicitly performs when training. Obtaining 
fine-tuned values of weights (W) and biases (b) such that the 
results from inputs (xi) during training matches the results of 
the control group leads to good neural network performance. 

f(xi,W) = W xi + b           (1) 

 

 

 

 

 
Figure 4 . Flowchart for preparing feedstock 

 

 

Step 1 

Water + Dispersant + 
Ceramic Powder 

 
  

Step 2 

Ball mill with ceramic 
beads 

 
  

Step 3 

Add binder to 
resulting slurry 

  

 

Step 4 

Mix with vacuum 
whip mixer 

 

  

Step 5 

Use vibratory table to 
set paste 



S. Choppala et al. /Future Technology                                                                                            May 2023| Volume 02 | Issue 02 | Pages 36-42 

40 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

A similar method is being implemented to further 
expand the efficiency and accuracy of the CODE process. De 
La Rosa [17] used transfer learning from a pre-trained VGG-
16 ML model using the Keras library with the Python 
programming language. Hyperparameters for this model 
were fine-tuned to work with newly generated data, and the 
overall image classification testing accuracy achieved was 
90%. The dataset used for training the model was generated 
with a compact USB camera capable of capturing high-
resolution images. Images were captured for defects after a 
random number of printed layers. Since CNNs, in general, 
require a large dataset for training to be able to predict with 
high accuracy while avoiding overfitting, dataset expansion, a 
form of data augmentation, was used by applying 
transformations to the images such as reshaping, rescaling, 
rotating, zooming, and altering brightness to already 
captured images. After applying dataset expansion, the 
images were then pre-processed by being normalized to be 
fed into the machine learning (ML) model. 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

By identifying the defects, the program updates the input 
G code to modify print parameters such as feed rate, nozzle 
temperature, material extrusion amount, and fan speed. 
These updates to the print settings serve as forms of solutions 
to fix common printing defects in real time. Extrusion-based 
ceramic 3D printing faces similar problems of common 
defects. A transfer learning approach to correctly identify 
defects within the CODE process is being applied. An initial 
stage of transfer learning would be to obtain a new database 
of images using a high-resolution camera, labels 
corresponding to different defects of CODE printing, and a 
control set of images and labels corresponding to no defects 
to perform training, validation, and testing. Supervised 
learning using the initial layer configuration and network 
hyperparameters is used to perform this process. The images 
generated could also be subject to data augmentation to 
increase the number of images for model training and 
validation. After studying the training dataset and test dataset 
results using both qualitative and quantitative analyses such 

  

Step 1 

Send green parts to 
humidity chamber 

   

Step 2 

Use furnace for 
debinding 

   

Step 3 

Use sintering furnace 
to densify parts 

Figure 6 . Geometrically complex zirconia (a), (b), (c) and alumina (d) specimens printed using CODE (Images (a), (b), and (c) 

reproduced from [20], (d) reproduced from [16] with permission from [Elsevier]) 

Figure 5 . Flowchart for postprocessing printed specimens 



S. Choppala et al. /Future Technology                                                                                            May 2023| Volume 02 | Issue 02 | Pages 36-42 

41 

 

as visual inspection and receiver operating characteristic 
(ROC) curves, respectively, the results are used to further fine 
tune and modify the neural network hyperparameters and 
layer configurations. The ROC curves are plots that are used 
to determine the frequency at which the CNN predicts the 
defect correctly and incorrectly. These curves plot the true 
positive rate on the x-axis and the false positive rate on the y-
axis and display the overall prediction accuracy of the 
network. The true positive and false positive rates are 
calculated using equations 2 and 3, respectively. The 
equations inputs include the frequency of true positive, false 
negative, false positive, and true negative based on the CNN 
performance. 

True Positive Rate = True Positive / (True Positive + False 
Negative)            (2) 

False Positive Rate = False Positive / (False Positive + True 
Negative)            (3) 

After training the neural network and obtaining a 
configuration such that the CODE printing defects are 
detected to a high level of accuracy and precision on the 
testing dataset, a control system feedback program could be 
implemented to update printing settings based on the defect 
detected in real-time. Since the CODE printer uses different 
hardware and has different printing mechanics than FFF, the 
program to update the G code would need to account for 
certain types of defects. CODE does not have certain 
mechanisms, such as a fan and a heated extruder nozzle, 
which, when controlled, is used for solutions of common 
defects in FFF. Instead, CODE uses a heat lamp, an oil bath, and 
a delay time between printing each layer. Therefore, the 
printing parameters needed to be updated with G-code are 
unique. Potential defects for CODE include stringing, over-
extrusion, under-extrusion, feedstock agglomeration, 
depletion of material, feedstock phase separation due to 
compressed air or material ratios, heat lamp distance, 
specimen drying time, and level of oil when a specimen is 
partially submerged. Potential solutions for the previously 
mentioned common defects for CODE include updating the G 
code to modify extrusion rate, nozzle travel speed, 
compressed air flow into the hopper with feedstock, the 
timing of the heat lamp, distance interval of the vertical axis, 
delay time between printing each layer. Furthermore, in some 
situations, the best way to further avoid common defects 
includes stopping the print. In cases of feedstock 
agglomeration, depletion of material, or feedstock phase 
separation, the print needs to be stopped to adjust the 
feedstock manually. 

8. Conclusions 

CODE is an extrusion-based 3D printing process for 
technical ceramics. The feedstock is in the form of a viscous 
paste. The main materials included in the feedstock are 
ceramic powder, deionized water, binder, and dispersant. 
This process has been tested on various technical ceramics, 
including zirconia and alumina. Parts with varying geometric 
complexity have been printed using the aforementioned 
materials. In order to have a completely printed part, there 
are three major phases: pre-processing, processing, and post-
processing. The paste will be prepared during the pre-
processing stages and printed during the processing stage. 
The gradual postprocessing is to ensure that the final ceramic 
specimens are fabricated without cracks, warpages, or 
gradients in mechanical properties. A 3D printer was 
specifically designed, fabricated, and controlled for the CODE 

process using LinuxCNC as the operating system. The 
electronics and physical hardware of the printer interface 
with the host computer using MESA boards which 
communicate signals through input G code on the LinuxCNC 
interface. A real-time CNN to detect common printing defects 
is being applied to the CODE process for defect detection as 
studied with FFF. Additionally, diverse data augmentation 
techniques are being applied to increase the overall dataset 
for the model to aid in defect detection performance. 
Furthermore, a feedback control program, built on the visual 
defect detection ML model, is being designed to automatically 
adjust the printing parameters in real time for implementing 
potential solutions corresponding with the common defects 
in the CODE process. This technique will promote overall 
printing accuracy and efficiency. 

Ethical issue 
The authors are aware of and comply with best practices in 
publication ethics, specifically with regard to authorship 
(avoidance of guest authorship), dual submission, 
manipulation of figures, competing interests, and compliance 
with policies on research ethics. The authors adhere to 
publication requirements that the submitted work is original 
and has not been published elsewhere. 

Data availability statement 
Data sharing is not applicable to this article as no datasets 

were generated or analyzed during the current study. 

Conflict of interest 

The authors declare no potential conflict of interest. 

References 

[1] I. Denry and J. Holloway, “Ceramics for dental 

applications: a review,” Materials, vol. 3, no. 1, pp. 351–

368, 2010. [Online] 

[2]  M. Bohner, L. Galea, and N. Doebelin, “Calcium 

phosphate bone graft substitutes: failures and hopes,” 

J. Eur. Ceram. Soc., vol. 32, no. 11, pp. 2663–2671, Aug. 

2012. [Online] 

[3]  F. Kamutzki, S. Schneider, J. Barowski, A. Gurlo, and D. 

A. H. Hanaor, “Silicate dielectric ceramics for 

millimetre wave applications,” J. Eur. Ceram. Soc., vol. 

41, no. 7, pp. 3879–3894, 2021. [Online] 

[4]  A. C. Young, O. O. Omatete, M. A. Janney, and P. A. 

Menchhofer, “Gelcasting of alumina,” J. Am. Ceram. 

Soc., vol. 74, no. 3, pp. 612–618, Mar. 1991. [Online] 

[5]  X. Deng, J. Wang, S. Du, F. Li, L. Lu, and H. Zhang, 

“Fabrication of porous ceramics by direct foaming,” 

Interceram, vol. 63, pp. 104–108, June. 2014. [Online] 

[6]  M. H. Bocanegra-Bernal, “Hot isostatic pressing (HIP) 

technology and its applications to metals and 

ceramics,” J. Mater. Sci., vol. 39, pp. 6399–6420, Nov. 

2004. [Online] 

[7]  H. Le Ferrand, “Magnetic slip casting for dense and 

textured ceramics: a review of current achievements 

and issues,” J. Eur. Ceram. Soc., vol. 41, no. 1, pp. 24–37, 

Jan. 2021. [Online] 

[8]  S. M. Olhero, P. M. C. Torres, J. Mesquita-Guimarães, J. 

Baltazar, J. Pinho-da-Cruz, and S. Gouveia, 

“Conventional versus additive manufacturing in the 

structural performance of dense alumina-zirconia 

ceramics: 20 years of research, challenges and future 

perspectives,” J. Manuf. Process., vol. 77, pp. 838–879, 

2022. [Online] 



S. Choppala et al. /Future Technology                                                                                            May 2023| Volume 02 | Issue 02 | Pages 36-42 

42 

 

[9]  A. Bove, F. Calignano, M. Galati, and L. Iuliano, 

“Photopolymerization of ceramic resins by 

stereolithography process: a review,” Appl. Sci., vol. 12, 

no. 7, p. 3591, Apr. 2022. [Online] 

[10]  M. A. Saadi, A. Maguire, N. T. Pottackal, M. S. Thakur, M. 

M. Ikram, A. J. Hart, P. M. Ajayan, and M. M. Rahman, 

“Direct ink writing: a 3D printing technology for 

diverse materials,” Adv. Mater., vol. 34, no. 28, p. 

2108855, Mar. 2022. [Online] 

[11]  A. Mostafaei, A. M. Elliott, J. E. Barnes, F. Li, W. Tan, C. L. 

Cramer, P. Nandwana, and M. Chmielus, “Binder jet 3D 

printing—process parameters, materials, properties, 

modeling, and challenges,” Prog. Mater. Sci., vol. 119, p. 

100707, June. 2021. [Online] 

[12]  N. Kamboj, A. Ressler, and I. Hussainova, “Bioactive 

ceramic scaffolds for bone tissue engineering by 

powder bed selective Laser Processing: a review,” 

Materials, vol. 14, no. 18, p. 5338, Sept. 2021. [Online] 

[13]  N. Travitzky, A. Bonet, B. Dermeik, T. Fey, I. Filbert-

Demut, L. Schlier, T. Schlordt, and P. Greil, “Additive 

manufacturing of ceramic-based materials,” Adv. Eng. 

Mater., vol. 16, no. 6, pp. 729–754, 2014. [Online] 

[14]  Z. Chen, Z. Li, J. Li, C. Liu, C. Lao, Y. Fu, C. Liu, Y. Li, P. 

Wang, and Y. He, “3D printing of ceramics: a review,” J. 

Eur. Ceram. Soc., vol. 39, pp. 661–687, 2019. [Online] 

[15]  A. Zocca, P. Colombo, C. M. Gomes, and J. Günster, 

“Additive manufacturing of ceramics: issues, 

potentialities, and opportunities,” J. Am. Ceram. Soc., 

vol. 98, no. 7, pp. 1983–2001, May 2015. [Online] 

[16]  A. Ghazanfari, W. Li, M. C. Leu, and G. E. Hilmas, “A 

novel freeform extrusion fabrication process for 

producing solid ceramic components with uniform 

layered radiation drying,” Addit. Manuf., vol. 15, pp. 

102–112, 2017. [Online] 

[17]  A. D. L. Rosa, “Defect detection and close-loop feedback 

using machine learning for fused filament fabrication,” 

San Jose State University, San Jose, CA, USA, 2022. 

[18]  A. Ghazanfari, W. Li, M. Leu, J. Watts, and G. Hilmas, 

“Mechanical characterization of parts produced by 

ceramic on-demand extrusion process,” Int. J. Appl. 

Ceram. Technol., vol. 14, no. 3, pp. 486–494, 2017. 

[Online] 

[19]  A. Ghazanfari, W. Li, M. C. Leu, J. L. Watts, and G. E. 

Hilmas, “Additive manufacturing and mechanical 

characterization of high density fully stabilized 

zirconia,” Ceram. Int., vol. 43, no. 8, pp. 6082–6088, 

2017. [Online] 

[20]  W. Li, A. Ghazanfari, D. McMillen, M. C. Leu, G. E. 

Hilmas, and J. Watts, “Characterization of zirconia 

specimens fabricated by ceramic on-demand 

extrusion,” Ceram. Int., vol. 44, no. 11, pp. 12245–

12252, 2018. [Online] 

 

 This article is an open-access article 

distributed under the terms and conditions of the Creative 

Commons Attribution (CC BY) license 

(https://creativecommons.org/licenses/by/4.0/). 


