See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/363794317

3D printing part orientation optimization: discrete approximation of support

volume

Article in International Journal of Electrical and Computer Engineering (IJECE) · December 2022
DOI: 10.11591/ijece.v12i6.pp5958-5966

CITATIONS READS

0 191

5 authors, including:

Juan Camilo Guachetá Alba Mauricio Mauledoux

Nueva Granada Military University Nueva Granada Military University
13 PUBLICATIONS 10 CITATIONS 111 PUBLICATIONS 175 CITATIONS

SEE PROFILE SEE PROFILE

Oscar Fernando Aviles

Nueva Granada Military University
184 PUBLICATIONS 407 CITATIONS

SEE PROFILE

All content following this page was uploaded by Juan Camilo Guachetá Alba on 23 September 2022.

The user has requested enhancement of the downloaded file.

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 12, No. 6, December 2022, pp. 5958~5966
ISSN: 2088-8708, DOI: 10.11591/ijece.v12i6.pp5958-5966  5958

3D printing part orientation optimization: discrete

approximation of support volume

Juan C. Guacheta Alba, Sebastian Gonzalez Garzon, Diego A. Nunez, Mauricio Mauledoux,
Oscar F. Aviles
DAVINCI Research Group, Mechatronics Engineering Department, Universidad Militar Nueva Granada, Colombia

Article Info ABSTRACT

Article history: In three-dimensional (3D) printing, due to the geometry of most parts, it is
necessary to use extra material to support the manufacturing process. This
Received Sep 13, 2021 material must be discarded after printing, so its reduction is essential to
Revised Jun 25, 2022 minimize manufacturing time and cost. An important parameter that must be
Accepted Jul 18, 2022 defined before starting the printing process is the part orientation, which has
repercussions on the quality, deposition path, and post-processing among
others. Usually, the user sets up this parameter arbitrarily, so this paper takes
Keywords: advantage of it on optimization techniques and proposes an approximation of
the volume be covered by the support material, which depends directly on
3D printing the angle of the part to be printed and its geometry. Among mono-objectives
Evolutionary operation optimization strategies, this work focuses on five of them. Their
Multivariable optimization performance is compared by two metrics: support volume and execution
Part orientation time. Then, the best result is compared with commercial software.
Random walk
This is an open access article under the CC BY-SA license.
Support volume

Corresponding Author:
Juan C. Guacheta Alba
DAVINCI Research Group, Mechatronics Engineering Department, Universidad Militar Nueva Granada
Carrera 11 # 101-80, Bogotá D. C., Colombia
Email: est.juan.guacheta@unimilitar.edu.co

1. INTRODUCTION
Additive manufacturing (AM), or more commonly known as 3D printing, consist of manufacturing
parts employing layer-by-layer material deposition, covering the solid volume. The movements made by the
printing nozzle are linked to the piece structure and orientation, and these determine the mechanical behavior
of the part and its texture and quality [1], [2]. The part's orientation is an initial parameter for its
manufacturing and is commonly set by the user. However, this variable is associated with characteristics of
the final product such as the model precision, the number of supports required, and the processing time for its
production [3], [4], as well as crucial end-user criteria, such as aesthetics, smoothness, material cost, and
energy spent in manufacturing [5]–[8]. Some issues related to wrong printing orientation are volumetric errors
that deform the part, high presence of the staircase effect represented in poor surface quality, high construction
time [4], material consumption [8], [9], anisotropy [10], and cylindricity and flatness errors [11]–[13].
3D printing has capability to manufacture any geometry compared to other manufacturing
processes. However, it is necessary for several parts to print support structures that guarantee structural
stability and avoid the collapse or deformation of the material in the regions with overhangs in the
manufacturing process [9], [14]. This support is eventually represented in waste material, additional costs
[15], and possible defects on the surfaces [9]. There are support materials that can be removed chemically,
improving the result of the part [14]. However, it is an additional process that affects the manufacturing time
and cost. This material is directly related to the support volume, which corresponds to the region used to
construct the holdup structure.

Journal homepage: http://ijece.iaescore.com

Int J Elec & Comp Eng ISSN: 2088-8708  5959

Some aspects such as: concavity, geometric shape, size, and islands must be considered to define
and obtain the support volume of the parts to be manufactured [16]. To treat and analyze the support volume,
it is crucial to consider a continuous but non-smooth function concerning the orientation angles [17]. Its
behavior will be defined by the geometry of the part to be manufactured. Several authors propose strategies
to obtain or approximate the support volume, such as the use of the kth nearest point algorithm [16], convex
hull surface triangles method [13], or a Quadtree decomposition [8] to find the volume of support structures.
Another essential aspect considered in the aforementioned strategies is the minimum self-support angle,
which is suggested for direct metal laser sintering (DMLS) printed parts of 45 degrees [14].
The above considerations have been a topic of study to optimize objective functions to improve the
printing process, being the principal applied function of the support volume. Some mono-objective
optimization techniques that have been applied to this problem are particle swarm optimization (PSO) [7],
ray-tracing method [13], optimization methods like electromagnetism [4], or perceptual models [9]. These
techniques become complex in their programming and implementation but guarantee to obtain global
solutions to the problem. Along with the support volume, multi-objective techniques have been applied to
optimize more objective functions, such as resolution error [5], [11], surface roughness [18], printing time,
and the number of aggregate suspensions [3]. The main multi-objective techniques used to solve this
optimization problem are genetic algorithms [3] and particle swarm optimizers [15].
Due to this, there is no evidence that focuses on optimizing the orientation of pieces that guarantees
a lower volume of supports. This work presents a solution using multivariate optimization techniques. This
paper is organized as follows: section 2 defines the method used to minimize the support material, employing
a correct orientation in printing pieces and multivariate optimization algorithms. Section 3 presents the
results obtained on a set of pieces. Also, the performance of the applied algorithms is compared. Finally,
section 4 points out the recommended algorithms in multivariate problems and exposes a discrete
approximation validation compared with commercial software. The following sections expose the capacity of
the discrete approximation of continuous problems and the feasibility of applying simple multivariate
optimization algorithms in problems with application in additive manufacturing, saving costs and time in
manufacturing pieces.

2. METHOD
The main objective of this work is to present an algorithm for optimizing the piece orientation. This
process requires several steps described below: first, it is required to define the decision variables that
correspond to the rotation angles of any piece in STL format. Second, the mathematical formulation and
programming of the objective function to be optimized, corresponding to the approximation of the volume
used to print the support material required in the printing. Third, the five multivariate optimization algorithms
used to solve the formulated problem are exposed as the performance metrics. At the end of this chapter, the
case study section presents pieces used to assess the algorithms and metrics.

2.1. Problem definition

To reduce the deposited material in the 3D printing process, the minimization of the support volume
of the parts, which is directly related to the deposited support material, is proposed. To structure the problem,
the three angles of rotation roll (ψ), pitch (θ) and yaw (φ), also called navigation angles or Tait-Bryan
angles, are defined as decision variables. These are necessary to obtain all possible orientations of the piece.
The rotation matrix used on the part, which represents a rotation in space, is the matrix R RPY defined in (1),
where 𝑅𝑧 , 𝑅𝑦 and 𝑅𝑥 are the matrix rotation at the 𝑧, 𝑦 and 𝑥 axis, respectively.

𝑅𝑅𝑃𝑌 = 𝑅𝑧,𝜑 𝑅𝑦,𝜃 𝑅𝑥,𝜓 (1)

With the three decision variables defined, the discrete approximation of the support volume is defined, which
corresponds to the objective function to be minimized. The discrete approximation of the support volume is
performed with the three defined decision variables, which corresponds to the objective function to be
minimized. When the part is obtained and the rotation matrix is applied, an octahedral mesh of thickness d is
created, completely covering the part. The mesh size affects the computing time and the precision of the
approximation of the support volume. The mesh is swept in the z-direction, and at each point it is evaluated
whether it corresponds to a region to be supported, considering the 45-degree rule for overhangs [14].
The pseudocode of this objective function is shown in Algorithm 1. Although it is possible with two
angles to find the orientations for printing, the discrete approach used to obtain the support volume is decided
to use the three angles. For selecting search methods and algorithms applied to the proposed problem, its
formal definition is made, which is an unconstrained single-objective multivariable optimization problem.
This problem is focused on its application on 3D printers and is mathematically represented in (2), where 𝑥
3D printing part orientation optimization: Discrete approximation … (Juan C. Guacheta Alba)
5960  ISSN: 2088-8708

corresponds to the angles roll (ψ), pitch (θ ) and yaw (φ ), while 𝑓(𝑥) corresponds to the approximate
support volume. Although the desired solution is sought in a range from -𝜋 to 𝜋, they are not constraints to
allow transitions between quadrants.

𝑚𝑖𝑛 𝑓(𝑥) ∴ 𝑥 = { 𝜑, 𝜃, 𝜓 } ∈ 𝑆 ⊆ ℝ3 (2)

Algorithm 1. Objective function pseudocode

Procedure Support volume approximation
Input: Piece; φ; θ; ψ; d. for each Point in meshing
Output: Volume. if Point is not inside the piece and Point
Rotation of the piece using matrix 𝑅𝑅𝑃𝑌 has no support at 45 degrees then
Create the mesh with thickness d
Point required for support
covering the piece
Location of mesh points inside the piece TotalPoints = sum(Necessary support points)
Z-direction mesh sweep Volume ← TotalPoints ∗ d 3

In Figure 1, the massive overhang test by Thingster is used to evaluate the performance of the
designed discrete objective function and its frame of reference is presented, as well as the rotation angles. It
can be observed that the volume covered with support material is not calculated for print angles less than 45
degrees. In contrast, for angles greater than this angle, this volume is approximated by the lower mesh. The
region of the mesh that approximates the support volume on the parts used in this paper is shown in green.

Figure 1. Discrete volume approximation of support material in massive overhang test

2.2. Optimization algorithms

In this section, five algorithms are executed. It is to analyze and compare solutions found and
evaluate the performance of multivariate algorithms. The algorithms used are described below.

2.2.1. Random walk

The random walk is a search method that describes a path that includes a succession of random steps
in the mathematical space [19]. This method generates a sequence of approximations by a unit random vector
generated at the i-th step. A random angle λu is generated if the i-th member of the group is chosen as the
wanderer of the z-th iteration [20]. Algorithm 2 exposes the pseudocode used for random walk scheduling,
defining parameters a step length, λ is set to 𝜋, a minimum allowable step length, ɛ is set to 1×10-4, and a
maximum allowable number of iterations, N is set to 20.

Algorithm 2. Random walk pseudocode

Procedure Random walk
Input: f(x); x0; λ; ɛ; N. else
Output: xmin if i<N then
f0 ← f(x0) i←i+1
i ← 0 goto 3
u ← random unit vector of n decision variables Else
x1 ← x0 + λ u λ←λ/2
f1 ← f(x) if λ≤ɛ then
if f1<f0 then xmin ← x0
x0 ← x1 exit;
f0 ← f1 Else
goto 2 goto 3

Int J Elec & Comp Eng, Vol. 12, No. 6, December 2022: 5958-5966
Int J Elec & Comp Eng ISSN: 2088-8708  5961

2.2.2. Evolutionary operation

The evolutionary operation (EVOP), introduced by George Box, is considered in applications where
only one output feature is optimized [21]. This method requires (2N+1) points, where N is the number of
decision variables, of which 2N are the corners of a hypercube. The (2N+1) function values are compared,
and the best point is identified [22]. If, in any iteration, the current point is not improved, then the size of the
hypercube is reduced. Algorithm 3 shows the pseudocode used for EVOP programming, defining as
parameters a tolerance ɛ→1×10-4, and a size reduction vector Δ→ [1,1,1] for the decision variables [φ, θ, ψ].

Algorithm 3. Evolutionary operation pseudocode

Procedure Evolutionary operation
Input: f(x); x0; Δ; ɛ; N. calculate f(x) in the (2N+1) points
Output: xmin 𝑥̅ ← point with the lowest function value
𝑥̅ ← x0
if ||Δ|| < ɛ then if 𝑥̅ = x0 then
xmin ← 𝑥̅ Δi ← Δi / 2
exit goto 2
else else
create 2N points by adding and x0 ← 𝑥̅
subtracting Δi/2 from each variable at goto 2
point 𝑥̅

2.2.3. Simplex method

It is the search method proposed by Spendley, Hext, and Himsworth and later improved by Nelder
and Mead, which seeks to minimize continuous and multidimensional unconstrained optimization problems
[23]. A regular simplex is a polyhedron composed of (N+1) equidistant points, which form its vertices, where
N is the number of decision variables. The main objective of the method is the generation of a new simplex
by projecting any vertex at an appropriate distance through some movements such as reflection (α),
expansion (γ), and contraction (β) [24]. Algorithm 4 details the operations and conditions necessary to apply
each of these movements. For the case study, the following parameters were used: α→0.6, γ→1.3, β→0.7
and a tolerance ɛ→100, related to the objective function in the scale of million.

Algorithm 4. Simplex method pseudocode

Procedure Simplex
Input: f(x); x0; α; γ >1; β ∈ (0,1); ɛ; N. else if f(xg)<f(xr)<f(xh) then
Output: xmin xnew ← (1+β)xc−βxh (contraction
Find xh (worst point), xl (best point), and xg outside)
(second worst point) Calculate f(xnew)
1
xc ← ∑N+1
i=1,i≠h 𝑥𝑖 (centroid) xh ← xnew
N
1
xr ← 2xc−xh (reflection) (𝑓(𝑥 )−𝑓(𝑥 ))2 2
Q ← [∑N+1
i=1
𝑖 𝑐
]
N+1
xnew ← xr if Q<ɛ then
if f(xr)<f(xl) then xmin ← xl
xnew ← (1+γ) xc − xh (expansion) exit
else if f(xr)≥f(xh) then else
xnew ← (1 −β) xc+β xh (contraction inside) goto 1

2.2.4. MATLAB fmincon function

As a comparison method for programming, the MATLAB fmincon function performs the search for
the minimum of a nonlinear multivariable scalar function, with or without restrictions. Like the three
previously mentioned, this method needs an initial point to be executed, so the same point is used for all
algorithm executions. The fmincon function allows the selection of five different algorithms, so the default
method interior-point was selected [25].

2.2.5. MATLAB ga function

Considering the algorithms investigated, the MATLAB ga function is used, which corresponds to a
genetic algorithm used to find the global minimum in highly nonlinear problems. In turn, it is based on a
natural selection process that mimics biological evolution and is applied on constrained and unconstrained
optimizations. Unlike classical algorithms, it generates a population of points for each iteration, and its
calculation uses random number generators [25].

2.3. Performance metrics

On each execution carried out, the following evaluation metrics are measured, which are grouped by
the algorithm for comparison:

3D printing part orientation optimization: Discrete approximation … (Juan C. Guacheta Alba)

5962  ISSN: 2088-8708

− Approximate part support volume: this metric corresponds to the objective function described in
Algorithm 1 and is expressed in mm3. A weighted average is performed on each implemented algorithm
for its evaluation, using (3), where x represents the approximate volume obtained for each part. In
contrast, w represents the inverse of the total volume of the part mesh, which is represented as a
percentage of deposited material for the mesh used.
− Algorithm execution time: this time is captured using a stopwatch timer and was taken only from the
algorithm, without considering the part reading or variable initialization, and is expressed in seconds. A
weighted average (𝑥̅ ) is performed using (3), where 𝑥 corresponds to the estimated time and weights 𝑤
corresponds to the inverse of the part mesh size.

∑ 𝑥𝑖 𝑤𝑖
𝑥̅ = ∑ 𝑤𝑖
(3)

2.4. Case study

A set of parts is selected to apply the objective function and algorithms mentioned. Figure 2 shows
the set of the parts used to evaluate the performance of the orientation optimization process, whose names
listed from left to right are: Wingnut_6x9 by Mike_mattala, Frog ring by Hlebushek2187, 3DBenchy by
Creative Tools, measuring cup 20 ml by Ndl, Samsung Galaxy Watch Stand by Lars_kglr, Simple bunny ball
joint doll by Dollightful, Smol Kitchen/Hobby Funnel by Towerdweller and Geo Cube by Burtronix. This
group collects pieces with simple and complex geometries, reflected in flat and curved surfaces and concave
and convex surfaces, to test the support volume estimation. All these pieces are licensed under the creative
commons licenses for use and sharing.
The mesh used to approximate the geometry has an average of 1.3 million points, being these cubes
with an average volume of 0.7 mm3. The programming and execution are performed in MATLAB R2021a
software. These tests were performed on a computer with an Intel core™ i5-10300H processor and installed
RAM of 8 GB. The multivariate optimization methods applied to the exposed geometries are the five
algorithms mentioned in the previous section, so eight executions were performed per algorithm.

Figure 2. Piece designs used for optimization

3. RESULTS AND DISCUSSION

Once the problem, the algorithms, and the case study to be used are formulated, the executions of
the five algorithms for the eight pieces are carried out. For each of them, the optimal orientation solution in
roll (ψ), pitch (θ ) and yaw (φ ) is obtained, as well as the measurement of the approximate support volume
of the piece and the execution time. The orientations obtained in all the executions are grouped to analyze
this data thoroughly. Table 1 shows the best solutions found by the optimization algorithms used. These
solutions are expressed in the range from -π to π, covering all possible orientations presented by the parts. It
is worth mentioning that, for the ‘Smol kitchen’ piece, the simplex method was the one that found the
optimal solution, while for the ‘Measuring cup’ and ‘Frog ring’ pieces, the optimal solution was found by the
random walk method. The evolutionary operation algorithm obtained the remaining optimal solutions, which
found the highest number of best solutions. The generic MATLAB algorithms were able to solve the
optimization problem correctly, but not with the precision of the first three.
Continuing with analyzing the optimal orientations obtained for the eight pieces, a visual evaluation
of the solutions is carried out. Figure 3 shows a set of with the optimal support volume approximations for
four selected parts. These solutions show that it solves orientations that visually can be selected for printing,

Int J Elec & Comp Eng, Vol. 12, No. 6, December 2022: 5958-5966
Int J Elec & Comp Eng ISSN: 2088-8708  5963

such as the ‘Frog ring’ piece exposed in Figure 3(a). In addition, it also correctly solves other parts, such as
the ‘Bunny's head’, shown in Figure 3(b), which has a hollow presence in its geometry, and large curved
sections, offering the user a solution that is not easy to obtain. In addition, in Figures 3(c) and 3(d), we can
see the prevalence of the 45-degree rule for printing, where a defined orientation for a flat face in these cases
results in greater volume and support material.

Table 1. Optimal orientation results for pieces

Pieces φ [rad] θ [rad] ψ [rad] Volume [mm3]
Wingnut 2.5743 0 -0.0012 3.5641
Frog ring 0.2076 3.1414 3.1415 7.8105
Smol kitchen 0.4524 0.0012 0.0273 1688.9
3Dbenchy 1.3750 0.5406 -0.4528 857.31
Samsung stand 0.5623 0.8740 0.0464 2468.4
Bunny's head -1.6258 -0.0005 -2.3223 879.028
Measuring cup -1.4447 2.3620 1.2969 370.012
Geo cube 0.2559 0 0.3341 1698.4

(a) (b) (c) (d)

Figure 3. Support volume in optimal piece orientations (a) frog ring, (b) Samsung stand, (c) measuring cup,
and (d) bunny's head.

To analyze the accuracy and speed of the algorithms, Figure 4 shows their performance curves in
the search for the orientation of the ‘Geo cube’ piece. The evolutionary operation and random walk
algorithms, on this optimization problem, have a faster convergence, also with a high accuracy on the first
iterations. Meanwhile, it is shown that the simplex algorithm finds with good accuracy the optimal solution,
but in a long time. On the other hand, the MATLAB fmincon function has low precision and the MATLAB
ga function consumes a high computation time per iteration.
Once the results related to the orientations obtained by the algorithms that most effectively
minimized the support volume have been analyzed, each algorithm's efficiency on all the pieces is analyzed.
For this, the evaluation metrics defined in section 2.4 are used: the approximate support volume of the part
and the algorithm time. It denotes that the weighted average is made using (3) on each algorithm for the set
of pieces. Table 2 shows the results of the metrics mentioned. These result from the weighted average for all
the pieces on each algorithm. The evolutionary operation search method found better solutions to the selected
parts for the approximate support volume metric, outperforming the simplex method and the MATLAB
genetic algorithm. The better result of the direct search methods is presented because the optimization
problem does not become complex and allows applying techniques based on the direct search. In addition,
being a discrete approximate function, the evolutionary operation algorithm facilitates its implementation and
execution.
On the other hand, the runtime metric for each algorithm performs better with MATLAB fmincon
function. However, these results are variant by part, as some are solved efficiently, while others do not
converge quickly. All algorithms consume a similar execution time except for the MATLAB genetic
algorithm, which takes 20 times longer due to its nature and the number of evaluations required on the
objective function. Finally, to corroborate the correct approximation of the volume to be coated required by
the support material, the open software PrusaSlicer is used, which exposes the printed material for the
manufacture of parts graphically. Figure 5 shows the approximation performed on the ‘3Dbenchy’ part in
Figure 5(a) and after, a support material by PrusaSlicer in Figure 5(b), presenting a similar result in both
cases. Because it is an approximation, the programmed objective function does not consider the necessary
support on the front holes of the part. However, it works correctly for orientation optimization, and the
discrete objective function is valid.

3D printing part orientation optimization: Discrete approximation … (Juan C. Guacheta Alba)

5964  ISSN: 2088-8708

Figure 4. Performance of the algorithms used on Geo cube

Table 2. Weighted results of volume and computation time metrics for all parts
Optimization techniques Volume [mm3] Time [s]
Random walk 135.5765 46.9499
Evolutionary operation 100.6915 45.1587
Simplex method 126.3512 44.5598
fmincon by MATLAB 194.9746 43.8417
GA by MATLAB 132.5944 947.53

(a) (b)

Figure 5. Support material used for ‘3Dbenchy’ printing (a) objective function and (b) PrusaSlicer

4. CONCLUSION
Direct search methods are not commonly used due to their rapid convergence into local solutions.
However, they can be applied to unrestricted problems with a finite search range and achieve good results.
Comparing the performance of the executed algorithms, it was observed that the random walk and
operational evolution methods were able to find the global minimum on the orientation optimization
problem. In contrast to this, the simplex method found local solutions, verified by the genetic algorithm used.
On the support volume optimization problem, the operational evolution method was the one that obtained the
best results, and with an execution time like that obtained by the MATLAB fmincon function. Simple search
methods were applied, which have a low computational cost compared to bio-inspired algorithms, which on
the case study analyzed, take up to 20 times longer. The operational evolution method converges on the same
global minimum as the genetic algorithm with higher accuracy.
The approximation of the support volume in a discrete way allows to calculate this variable to any
piece, regardless of its geometry, making this process automatic and simple, to comparison of its
mathematical formulation. The effectiveness of this approach is corroborated by a commercial 3D printing
software, where the regions needed for support printing correspond to the discrete support volume. Unlike
other research, non-symmetric geometries such as ‘Samsung stand’ are used, so the search surface is more
extensive and presents a unique global minimum. Finally, it is recommended to use optimization strategies in
the preprocessing phase in 3D printing, to select the best orientation to parts in which it is not possible
visually, allowing to reduce costs and time.

Int J Elec & Comp Eng, Vol. 12, No. 6, December 2022: 5958-5966
Int J Elec & Comp Eng ISSN: 2088-8708  5965

Since the present proposed algorithm automatically selects the optimal print orientation, it seeks to
make a preprocessing software that offers the best part orientation to the user before printing his part. This
process ensures that it consumes less printing material, which is reflected in the cost and contamination in the
manufacturing process. In addition, the use of GPUs to parallelize the discrete objective function is proposed
since its computation time can be reduced with this strategy. Finally, it would be interesting to apply multi-
objective optimization techniques having more objective functions, such as the staircase effect, printing time
or surface finish, and other decision variables such as the printing pattern.

ACKNOWLEDGEMENTS
This paper is a research product from the IMP-ING-3122 high-impact project funded by the
Research Vice-Chancellor of Universidad Militar Nueva Granada–2021. Producto derivado del proyecto
IMP-ING-3122 financiado por la Vicerrectoría de Investigaciones de la Universidad Militar Nueva Granada
–Vigencia 2021.

REFERENCES
[1] L. Di Angelo, P. Di Stefano, and A. Marzola, “Surface quality prediction in FDM additive manufacturing,” The
International Journal of Advanced Manufacturing Technology, vol. 93, no. 9–12, pp. 3655–3662, Dec. 2017, doi:
10.1007/s00170-017-0763-6.
[2] J. Mueller and K. Shea, “The effect of build orientation on the mechanical properties in inkjet 3D-printing,” in Twenty-Sixth
Annual International Solid Freeform Fabrication (SFF) Symposium – An Additive Manufacturing Conference, Austin, United
States, 2015, pp. 983–992.
[3] Z. Li, A. Hu, J. Fu, X. Wu, and H. Li, “Printing orientation optimization of 3D model,” in Proceedings of the 2nd International
Conference on Computer Science and Application Engineering - CSAE ’18, 2018, pp. 1–5, doi: 10.1145/3207677.3278034.
[4] M. A. Matos, A. M. A. C. Rocha, and A. I. Pereira, “Improving additive manufacturing performance by build orientation
optimization,” The International Journal of Advanced Manufacturing Technology, vol. 107, no. 5–6, pp. 1993–2005, Mar. 2020,
doi: 10.1007/s00170-020-04942-6.
[5] J. Son and S. Choi, “Orientation selection for printing 3D models,” in 2015 International Conference on 3D Imaging (IC3D),
Dec. 2015, pp. 1–6, doi: 10.1109/IC3D.2015.7391823.
[6] W. M. Wang, C. Zanni, and L. Kobbelt, “Improved surface quality in 3D printing by optimizing the printing direction,” Computer
Graphics Forum, vol. 35, no. 2, pp. 59–70, May 2016, doi: 10.1111/cgf.12811.
[7] M. Barclift, A. Armstrong, T. W. Simpson, and S. B. Joshi, “CAD-integrated cost estimation and build orientation optimization to
support design for metal additive manufacturing,” Aug. 2017, doi: 10.1115/DETC2017-68376.
[8] P. Das, R. Chandran, R. Samant, and S. Anand, “Optimum part build orientation in additive manufacturing for minimizing part
errors and support structures,” Procedia Manufacturing, vol. 1, pp. 343–354, 2015, doi: 10.1016/j.promfg.2015.09.041.
[9] X. Zhang, X. Le, A. Panotopoulou, E. Whiting, and C. C. L. Wang, “Perceptual models of preference in 3D printing direction,”
ACM Transactions on Graphics, vol. 34, no. 6, pp. 1–12, Nov. 2015, doi: 10.1145/2816795.2818121.
[10] J. F. P. Lovo, C. A. Fortulan, and M. M. da Silva, “Optimal deposition orientation in fused deposition modeling for maximizing
the strength of three-dimensional printed truss-like structures,” Proceedings of the Institution of Mechanical Engineers, Part B:
Journal of Engineering Manufacture, vol. 233, no. 4, pp. 1206–1215, Mar. 2019, doi: 10.1177/0954405418774603.
[11] R. Paul and S. Anand, “Optimization of layered manufacturing process for reducing form errors with minimal support structures,”
Journal of Manufacturing Systems, vol. 36, pp. 231–243, Jul. 2015, doi: 10.1016/j.jmsy.2014.06.014.
[12] G. Moroni, W. P. Syam, and S. Petrò, “Functionality-based part orientation for additive manufacturing,” Procedia CIRP, vol. 36,
pp. 217–222, 2015, doi: 10.1016/j.procir.2015.01.015.
[13] M. P. Zwier and W. W. Wits, “Design for additive manufacturing: Automated build orientation selection and optimization,”
Procedia CIRP, vol. 55, pp. 128–133, 2016, doi: 10.1016/j.procir.2016.08.040.
[14] A. T. Gaynor and J. K. Guest, “Topology optimization considering overhang constraints: Eliminating sacrificial support material
in additive manufacturing through design,” Structural and Multidisciplinary Optimization, vol. 54, no. 5, pp. 1157–1172, Nov.
2016, doi: 10.1007/s00158-016-1551-x.
[15] H. Shen, X. Ye, G. Xu, L. Zhang, J. Qian, and J. Fu, “3D printing build orientation optimization for flexible support platform,”
Rapid Prototyping Journal, vol. 26, no. 1, pp. 59–72, Jan. 2020, doi: 10.1108/RPJ-09-2018-0252.
[16] M. A. Habib and B. Khoda, “Attribute driven process architecture for additive manufacturing,” Robotics and Computer-
Integrated Manufacturing, vol. 44, pp. 253–265, Apr. 2017, doi: 10.1016/j.rcim.2016.10.003.
[17] B. Ezair, F. Massarwi, and G. Elber, “Orientation analysis of 3D objects toward minimal support volume in 3D-printing,”
Computers & Graphics, vol. 51, pp. 117–124, Oct. 2015, doi: 10.1016/j.cag.2015.05.009.
[18] S. E. Brika, Y. F. Zhao, M. Brochu, and J. Mezzetta, “Multi-objective build orientation optimization for powder bed fusion by
laser,” Journal of Manufacturing Science and Engineering, vol. 139, no. 11, Nov. 2017, doi: 10.1115/1.4037570.
[19] F. Xia, J. Liu, H. Nie, Y. Fu, L. Wan, and X. Kong, “Random walks: A review of algorithms and applications,” IEEE
Transactions on Emerging Topics in Computational Intelligence, vol. 4, no. 2, pp. 95–107, Apr. 2020, doi:
10.1109/TETCI.2019.2952908.
[20] Z. Li and Z. Li, “A slicing parameter optimization method using group search optimization algorithm in STL model for 3D
printing application,” Academic Journal of Engineering and Technology Science, vol. 2, no. 3, pp. 73–84, 2019, doi:
10.25236/AJETS.020057.
[21] K. Rutten, J. De Baerdemaeker, J. Stoev, M. Witters, and B. De Ketelaere, “Constrained online optimization using evolutionary
operation: A case study about energy-optimal robot control,” Quality and Reliability Engineering International, vol. 31, no. 6,
pp. 1079–1088, Oct. 2015, doi: 10.1002/qre.1662.
[22] C. A. C. Coello, Introduction to evolutionary computing, (in Spanish), 2nd ed. Computación Evolutiva, Mexico: Academia
Mexicana de Computación, 2019.
[23] S. Zapotecas-Martínez and C. A. C. Coello, “MONSS: A multi-objective nonlinear simplex search approach,” Engineering
Optimization, vol. 48, no. 1, pp. 16–38, Jan. 2016, doi: 10.1080/0305215X.2014.992889.

3D printing part orientation optimization: Discrete approximation … (Juan C. Guacheta Alba)

 5966  ISSN: 2088-8708

[24] S. Z. Martinez, A. A. Montano, and C. A. C. Coello, “A nonlinear simplex search approach for multi-objective optimization,” in
2011 IEEE Congress of Evolutionary Computation (CEC), Jun. 2011, pp. 2367–2374, doi: 10.1109/CEC.2011.5949910.
[25] I. The MathWorks, “Optimization toolboxTM user’s guide,” 2021. https://www.mathworks.com (accessed Jul. 23, 2021).

BIOGRAPHIES OF AUTHORS

Juan C. Guacheta Alba received his B.S. in mechatronics engineering from

Universidad Militar Nueva Granada in 2020. Currently, he is working toward the M.Sc. degree
in Mechatronics Engineering at Universidad Militar Nueva Granada. His research interests
include robotics, multi-agent systems and optimization. He can be contacted at email:
est.juan.guacheta@unimilitar.edu.co.

Sebastian Gonzalez Garzon received his B.S. in mechatronics engineering from

Universidad Militar Nueva Granada in 2020. Currently, he is working toward the M.Sc.
degree in Mechatronics Engineering at Universidad Militar Nueva Granada, Colombia. His
research interests include optimization and robotics. He can be contacted at email:
est.sebastian.gonz1@unimilitar.edu.co.

Diego A. Nunez received his B.S. in mechatronics engineering from Universidad

Militar Nueva Granada in 2005 and his M.Sc. in mechanical engineering from Universidad de
los Andes in 2014. He is specialist in polymer processing and is currently working toward a
Ph.D. in applied science at Universidad Militar Nueva Granada. His research interests include
parallel robots, optimization, and additive manufacturing. He can be contacted at email:
danvmoldes@gmail.com.

Mauricio Mauledoux received his B.S. in mechatronics engineering from

Universidad Militar Nueva Granada, in 2005. In 2011, he received his PhD degree in
Mathematical models, numerical methods, and software systems (Red Diploma) from St.
Petersburg State Polytechnic University, Russia. In 2012, he joined the Department of
Mechatronic Engineering, at Universidad Militar Nueva Granada, in Colombia as Assistant
Professor. His current research interests include robotics, automatic control, multi-agent
systems, smart grids, and optimization. He can be contacted at email:
mauricio.mauledoux@unimilitar.edu.co.

Oscar F. Aviles received his B.S. in electronics engineering from Universidad

Antonio Nariño, in 1995. He received his PhD degree in Mechanical Engineering (Robotics
Devices) in 2008 from Universidade Estadual de Campinas, Brazil. In 1998, he joined the
Department of Mechatronics Engineering at Universidad Militar Nueva Granada, Colombia.
His current research interests include robotics and biomechatronic. He can be contacted at
email: oscar.aviles@unimilitar.edu.co.

Int J Elec & Comp Eng, Vol. 12, No. 6, December 2022: 5958-5966

View publication stats