Optimization of Lattice Structures For Additive Manufacturing Technologies
Optimization of Lattice Structures For Additive Manufacturing Technologies
1 Introduction
for aerospace industry). To reach this goal, commercial CAD software applica-
tions exist, that are able to create a skin model and an internal lattice structure.
Unfortunately it is very difficult to perform structural analysis on cellular geomet-
ric models. Alternatively, other commercial tools support topology optimization,
which modifies material layout within a given design space, for a given set of
loads and boundary conditions such that the resulting layout meets a prescribed set
of performance targets, obtaining an optimized concept design.
Today, interest in cellular materials is being driven by transport industry, aimed
at new vehicles, which need to be lighter than ever (to reduce fuel usage and iner-
tia) but also stiff, strong and capable of absorbing mechanical energy (e.g. in vehi-
cle collision or in helmet design) [2-3]. This explains the number of papers dealing
with homogenous lattice structures and related mechanical properties. Otherwise
conformal or random cellular structures were studied in literature and optimization
criteria were proposed. For instance, recent research proposed methods for opti-
mizing cellular structures, where the goal is to reach an established deflection and
a target volume, ensuring structural strength [4]. The approach was extended to
conformal lattice structures, in which the cellular structures are not regular, but
follow the shape of curved surfaces in order to increase its stiffness or strength [5].
Another optimization method of conformal lattice structure use the relative
density available from the topology optimization to assign a thickness to the
beams [6]. A Bidirectional Evolutionary Structural Optimization approach based
on topology optimization was recently proposed. This method takes into account
the orientation of cells in the design stage, and considers solid volume and skin in
addition to beam elements [7].
In a recent work the authors [8] proposed a Computer-Aided method for gener-
ative design and optimization of regular cellular structures, obtained by repeating
a unit cell inside a volume, where the elements are cylinders having different radii.
The approach is based on the iterative variation of the radius of each element in
order to obtain the optimal design. Target of the optimization is the achievement
of a required level of utilization, that specifies the level of usage of the material
for each element (utilization is equal to zero when the maximum stress inside an
element is null and is equal to 1 when the maximum stress is the maximum admis-
sible, e.g. equal to yield stress).
The method was investigated in few configurations of a cantilever beam, con-
sidering six different cell types and two load conditions. As a result, cell types
were classified as a function of the relative density and compliance/stiffness in the
different load conditions. The main limit of the study concerns the limited number
of tests performed and the absence of case studies and experimental tests.
In this work, a number of test cases has been assessed in order to evaluate the
behavior of the method during the iterations, and the effects of different loads and
cell dimensions. These results will be the basis for the development of guidelines
in parameters setup as a function of load/constraint configuration and compli-
ance/stiffness requirements. Finally, a visual comparison between the proposed
method and topology optimization approach is shown.
Optimization of lattice structures … 215
2 Design Method
The proposed design method (fig. 1) is aimed at the substitution of a solid model
with cellular structures, obtaining a wire model computed by a generative model-
ing approach [9]. A finite element (FE) model is built on the wire model and then
analyzed [10]. A dedicated iterative optimization procedure was developed in Py-
thon [11] in order to obtain an optimized geometric model.
Repeating side by side a regular unit cell of specified dimension, a wire model
is obtained. Each type of unit cell is defined by a number of edges and conse-
quently the wire model is a collection of lines connected at vertices called nodes.
Each edge of the wire model is a beam with circular sections in the FE model.
The initial radius is the same for all the beams, and is computed in order to ensure
a desired value of utilization index for the most stressed beam. This index speci-
fies the level of usage of the material for an element according to EN 1993-1-1
[12]. To complete the FE model, material, load and constraints must be defined
according to functional requirements of the solid model.
The most important result of FE analysis is the computation of the utilization of
each beam (Ui = utilization of i-th beam), needed in the optimization step. Goal of
the optimization is to obtain Ui of all beams close to a target utilization Ut. In or-
der to consider the AM process features, a minimum radius (Rmin) for each beam
must be defined; moreover a max radius (Rmax) is computed considering the cell
dimension.
More in detail, the optimization procedure consists of an iterative modification
of the radius Ri of each beam (therefore defining a new FE model) and involves
new results of the FE analysis. Each new radius Rn i is defined as:
Rni Ri Ui Ut (1)
The iterative procedure continues until Ui of each beam satisfies the following
equation:
at the nodes, where several beams having different radii converge together. A sim-
ilar approach was proposed by Wang et al. [13].
This modeling procedure shows limits especially in Boolean operations, file
dimensions and fillets. To overcome these restrictions a specific modeling proce-
dure was developed for cubic cell. Starting from the results of the optimization
procedure, a simple mesh was modeled and then the Catmull-Clark subdivision
surface [14] was adopted to obtain a smooth mesh using Weaverbird [15]. This
approach can be extended to other cell types, defining specific methods for creat-
ing a simple mesh model of the cell.
Solid model
Cell type
Cross-sections
Material FE model
Loads
Constrains FE analysis
N
Optimized? New radii
Y
Optimized
model
Mesh
model
Fig. 1. The proposed method for modeling and optimize lattice structures.
3 Test cases
a) b) c) d) e) f)
Fig. 2. Cell types: a) SC, b) BCC, c) RBCC, d) OT, e) GAM, f) WG.
The behavior of the method during the iterations has been investigated on the 6
cell type, adopting 5 mm of cell dimension and 50 N of flexural load.
The effect of the load has been studied on a 5 mm BCC subjected to a flexural
load ranging between 10 N to 200 N, with step of 10 N. The cell dimension effect
has been investigated on a BCC cell with edge length 2.5 mm,5 mm,10 mm.
Comparison between our method and topological optimization has been per-
formed on SC cell with edge length 2.5 mm and 5 mm on 50 N of flexural load.
The topology optimization problem has been solved using Millipede, an add-on
for Grasshopper [20].
The convergence conditions adopted are: Ut=0.5, x=0.10 (0.45<U i<0.55), Rmin
= 0.25 mm, Rmax = 5 mm.
Relative density U is assumed as:
ρ=Vo/Vc (5)
where Vo is the volume of the optimized structure and Vc is the volume of the
cantilever (Vc = 30x30x80 = 72000 mm3). Volume has been computed without
considering the beams ends overlapping (i.e. without performing any boolean un-
ion).
The behavior of the method during the iterations is shown in fig. 3 and summa-
rized in tab. 1 for the convergence conditions. The maximum and minimum utili-
zations of the beams show that convergence can be obtained with a low number of
iterations for the BCC and RBCC cell. These 2 cells show a clear trend in conver-
gence, while the other cells have an irregular trend in the maximum and minimum
utilizations values and consequently on the method convergence (fig. 3a). Number
of beams and nodes show the problem complexity: the simplest cells are CS, WG
and BCC (tab. 1). BCC shows the lowest relative density in the studied conditions,
218 G. Savio et al.
while the GAM the highest (fig. 3b). It should be underlined that the relative den-
sity has a contribution linked to the beams with minimum radius, and consequent-
ly, in other configurations, different types of cell may produce lower relative den-
sity. WG, RBCC and BCC show the higher stiffness, while the GAM has the
higher compliance (fig 3c).
Generally it is possible to see that higher density is related to a lower compli-
ance in the same topological configuration (fig 3b,c). This aspect is evident for the
WG cell, in which it is possible to see 2 configurations: the first around 20th itera-
tion and the second beyond 60th iteration.
Other convergence criteria could be adopted in order to obtain a lower number
of iterations, with no significant difference in relative density and displacement.
For example, using as convergence criteria Ui<0.55 and a variation of displace-
ment between two consecutive iterations less than 0.1%, the GAM cell converges
within 12 iterations. Similarly, using as convergence criteria Ui<0.55 and a varia-
tion of relative density between two consecutive iterations less than 0.1%, the WG
cell converges within 24 iterations.
Results relevant to the load variation, studied on a BCC cell, are summarized in
fig. 4. The proposed method found a solution until a load of 140 N. In order to in-
crease the maximum load value, it is possible to change the convergence criteria
or modify the approach for the new radii computation. In this range the relative
density is almost proportional to the load, while the displacement has a quadratic
behavior for load ranging between 20 and 140N. The number of iterations for the
convergence is between 16 and 25 for loads less than or equal to 120 N and in-
crease until 40-50 iterations for higher loads.
Adopting different cell dimensions (fig. 5), higher stiffness could be obtained
with smaller cell dimensions. This could be related to the increased number of
beams with minimum radius. Due to the same reason, increasing the cell dimen-
sion, the relative density shows a decreasing trend. For the given conditions the it-
erations needed increase together with the cell dimension. Finally, it is possible to
see a strong reduction of the problem complexity (number of beams) increasing
the cell dimension.
Fig. 6 shows a visual comparison between our approach and topology optimi-
zation. A similar behavior can be seen particularly in the portion close to the con-
strain (left) and to the loaded surface (right). These similarities could be related to
the homogenization procedure that occurred in the topology optimization problem
[21], in which the design space is filled with an artificial composite material made
of cells with holes.
In brief, the results shown in this paper and in [8] can be used to derive guide-
lines in cell selection and parameters setup: when the stiffness is the design target,
RBCC and BCC cells structures are recommended. CS shows the lowest complex-
ity. Increasing the load, the relative density and displacement increase. Reducing
the cell dimension, the relative density, stiffness and number of beams increase.
BCC are suggested when the goal is a low relative density and low iterations.
Higher values of the range of admissible utilization allow faster convergence.
Optimization of lattice structures … 219
1
CS
BCC
GAM
0.75 WG
OT
RBCC
Utilization
0.5
0.25
0
0 20 40 60 80 100
a) Iterations
0.12
0.1
CS BCC GAM
Relative density
WG OT RBCC
0.08
0.06
0.04
0 20 40 60 80 100
b) Iterations
c)
Fig. 3. Behavior of the proposed method: a) utilization, b) relative density and c) displacement as
a function of the iterations under flexural load.
220 G. Savio et al.
0.1 4
Displacement [mm]
Relative density
0.075 3
0.05
Relative 2
density
0.025 1
0 50 Load [N] 100 150
Fig. 4. Relative density as a function of the flexural load for the BCC cell.
0.1 5
0.08 4
Displacement [mm]
Relative density
0.06 3
0.04 2
Relative density
Displacement
0.02 1
0 0
0 Cell dimension [mm] 5 10
a)
60000 30
50000 25
Number of beams
40000 20
Iterations
30000 15
Number of Beams
Iterations
20000 10
10000 5
0 0
0 Cell dimension [mm] 5 10
b)
Fig. 5. Behavior of the proposed method: a) relative density and displacement and b) number of
beams and iterations as a function of the cell dimension under flexural load for the BCC cell.
a)
b)
c)
Fig. 6. Proposed method on a cubic cell (a,b), vs topology optimization (c).
222 G. Savio et al.
References