Accepted Manuscript

Modelling and characterization of a porosity graded lattice

structure for additively manufactured biomaterials

Mathieu Dumas, Patrick Terriault, Vladimir Brailovski

PII: S0264-1275(17)30149-1
DOI: doi: 10.1016/j.matdes.2017.02.021
Reference: JMADE 2761
To appear in: Materials & Design
Received date: 10 December 2016
Revised date: 27 January 2017
Accepted date: 8 February 2017

Please cite this article as: Mathieu Dumas, Patrick Terriault, Vladimir Brailovski ,
Modelling and characterization of a porosity graded lattice structure for additively
manufactured biomaterials. The address for the corresponding author was captured as
affiliation for all authors. Please check if appropriate. Jmade(2017), doi: 10.1016/
j.matdes.2017.02.021

This is a PDF file of an unedited manuscript that has been accepted for publication. As
a service to our customers we are providing this early version of the manuscript. The
manuscript will undergo copyediting, typesetting, and review of the resulting proof before
it is published in its final form. Please note that during the production process errors may
be discovered which could affect the content, and all legal disclaimers that apply to the
journal pertain.
 ACCEPTED MANUSCRIPT

Modelling and characterization of a porosity graded lattice structure for additively

manufactured biomaterials

Mathieu Dumas1, a, Patrick Terriault1, b, Vladimir Brailovski1, c

1
Ecole de technologie superieure, Department of Mechanical Engineering, 1100 Notre-Dame
Street West, Montreal, Quebec H3C1K3, Canada
a
mathieu.dumas.1@etsmtl.net, bpatrick.terriault@etsmtl.ca, cvladimir.brailovski@etsmtl.ca

Keywords: Porous material, Biomaterial, Additive manufacturing, Diamond-type unit cell,

PT
Finite element analysis

Abstract. A novel approach has been proposed for the design of porosity-graded lattice

RI
structures suitable for bone replacement applications. A MATLAB routine was developed in order
to model a lattice structure and generate an STL file for additive manufacturing. With this routine,

SC
geometric properties, such as the pore size and strut thickness, can be controlled to provide the
desired porosity distribution and mechanical properties of the structure.

The stiffness and yield stress of three titanium structures with densities of 20, 42 and 60% were
NU
evaluated by finite element analysis and tensile testing. The experimental results diverged by up to
50% from those predicted by the finite element analysis. The geometrical deviations between the
modelled and fabricated structures and the approximations made for the material modelling were
MA

considered to be at the root of these discrepancies.

The experimentally evaluated Young’s modulus ranged from 1.6 to 20.3 GPa for the 20 and 60%
dense structures, respectively, while the yield stress ranged from 23 to 181 MPa for the same
ED

structures. These results are comparable to those for cortical and trabecular bone. Scaling relations
for the stiffness and yield stress of porous structures were developed to allow the computer-assisted
design of porosity-graded load-bearing implants.
PT

INTRODUCTION

Current orthopedic implants are generally made of titanium and chromium alloys or stainless
CE

steel which are relatively stiff materials when compared to bone. This stiffness mismatch can
generate stress shielding between the implant and surrounding bone and lead to premature implant
loosening and implantation failure. Porous materials with controlled stiffness can be used to
AC

improve the mechanical compatibility of implants with bone tissue. In addition to increasing
flexibility, this porosity can help with implant fixation by allowing bone ingrowth [1]. The design
and fabrication of these porous materials can however be a major challenge because both
mechanical and biological criteria must be considered.

To ensure mechanical compatibility at the implantation site, implants should mimic properties of
bone, but the latter vary depending on several factors, such as age, sex and anatomical site. For
example, bone tissue stiffness can vary from up to 20 GPa in cortical bone to well below 1 GPa in
cancellous bone [2]. Bones can also have a graded structure, such as in regions between the soft and
hard tissue of bone-to-cartilage interfaces [2]. In order to match the mechanical and biological
properties of bone, functionally graded materials must be designed.

Several manufacturing methods, such as powder sintering and pressurized pore expansion, have
been used to create metallic materials with controlled porosity [3-5]. These processes have however
mostly been used to generate materials with a random pore distribution. With developments in
additive manufacturing (AM), many porous implants for load-bearing applications are being
 ACCEPTED MANUSCRIPT

developed [6-9]. More specifically, powder bed fusion processes have attracted significant attention
because of the complexity of the components they can produce. Given the fine-resolution features
that can be obtained, AM allows for the fabrication of ordered truss-like arrangements called lattice
structures [10]. These structures have been shown to generally have a more uniform stress
distribution and better long-term mechanical response than stochastic structures [11].

Currently, one of the most common approaches for lattice structure design involves the use of
computer-aided design (CAD) software [12, 13]. Typically, a solid unit cell structure is designed or
taken from a unit cell library and copied to generate the lattice structure. However, when modeling
large-scale heterogeneous structures, this approach requires considerable computer resources [14,
15]. Additional computational resources can also be necessary when generating an STL file from an

PT
original CAD file, the STL file being the typical format for transferring models within the AM
workflow.

RI
Several alternatives to CAD-based modelling have been studied in order to reduce memory
requirements and ease the lattice structure design process. Hollister [16] proposed an image-based
design approach for creating lattice structures with tailored properties by using patient-specific

SC
medical imaging data. Fryazinov, et al. [17] and Yoo [18] proposed the use of implicit functions to
model periodic structures in a compact form. In this case, the parameterization of the modelled
structures becomes relatively straightforward, requiring simple adjustments of the underlying
NU
functions. The main disadvantage of these design approaches is that they require additional
processing to convert the modelled structure into an STL file to allow AM fabrication, which
reduces the efficiency of the design workflow.
MA

Mmillan, et al. [19] and Armillotta and Pelzer [20] used different approaches to create tessellated
surfaces of lattice structures, which could be used directly to generate STL files. Both approaches
consist in populating a uniform grid with tessellated unit cells. While the approach proposed by
Mmillan, et al. [19] allows for the selection of different unit cell types, Armillotta and Pelzer [20]
ED

used rectilinear struts aligned with the grid to create a Cartesian pore network. They then went on to
apply their strategy to create the lattice structure within a given spatial region, and discussed how to
generate such a structure with a graded pore size. This strategy, however, required the
PT

implementation of a carefully planned parametric algorithm in order to create the desired gradient.
Although the approach has significant advantages, a tool with better control over porosity gradients
needs to be developed.
CE

Specialized software applications have begun to incorporate lattice structure design tools which
can help create intricate graded structures. Some such tools could potentially be used to evaluate the
mechanical properties of these structures and optimize their configuration, depending on the loading
AC

conditions and functional requirements. This optimization is of particular interest for reducing stress
shielding-related phenomena, while ensuring implant resistance. To the best of the authors’
knowledge, however, none of the commercially available software allows a fully integrated and
efficient workflow comprising the design, optimization and potential fabrication of lattice structures
via additive manufacturing.

One of the approaches used to simulate the global behaviour of porous implants under given
loading conditions involves the use of multi-scale finite element (FE) methods [21]. Multi-scale
modelling is necessary with this approach because the computing resources required to model each
structural element at the scale of an implant are too high to be provided with even high-performance
clusters. With multi-scale modeling, when analyzing the global behaviour of an implant, the
apparent properties of a representative volume element (RVE) can be evaluated at a mesoscale and
transferred to a macroscale model. To this end, Gibson and Ashby [22] developed simple scaling
relations which can be used to tie the relative density of porous structures to their apparent
mechanical properties, such as stiffness and strength.
 ACCEPTED MANUSCRIPT

Once mesoscale relations have been derived, different approaches may be used to evaluate the
properties of graded structures at the macroscale. Simoneau, et al. [23] defined an RVE to make a
bridge between the meso- and macroscales. Van Grunsven, et al. [24] proposed the use of the rule
of mixtures to predict the elastic behaviour at the macroscale by considering graded lattice
structures as a series of uniform layers of equal thickness. Other authors also suggested the use of
various homogenization techniques to evaluate the mechanical behavior of graded structures [25-
27].

The objective of this work was twofold. First, an efficient approach for modelling multi-scale
lattice structures for biomedical applications had to be developed. The approach needed to ease the
lattice structure design process for additive manufacturing, specifically for powder bed fusion, and

PT
to be capable of being easily integrated within an optimization algorithm. The specific requirements
and the approach which were developed to meet this objective will be discussed herein. The second
objective was to evaluate the mechanical properties of the designed structures in order to determine

RI
relations which could be used for FE implant simulations and FE-based implant design. Tensile
testing was therefore needed to evaluate the apparent properties of the structures and to compare the
experimental results with numerical predictions.

SC
MATERIALS AND METHODS NU
Lattice structure design. The current work focuses on the development of an approach for
modelling, via an original MATLAB routine, functionally graded lattice structures potentially
suitable for bone replacement applications. In order to achieve this goal, emphasis is laid on the
following mandatory steps:
MA

 Model lattice structures which respect the bone ingrowth criteria

 Generate structures with easily defined porosity gradients

ED

 Confine the structure created within a given design space

 Generate a valid STL file for AM

PT

In the proposed approach, which is based on the work of Armillotta and Pelzer [20], a
CE

voxelization function (Atkinson, 2016) is used to generate a uniform grid within a given design
space. The function is used to convert a polygon mesh from an STL file input and generate a
volume representation. The volume representation consists of cubic cells, which are either filled
with material or are void. A uniform grid is generated by connecting the edges of the filled cubes
AC

together. The grid can ultimately be populated by unit cells. It should be noted, however, that
although an STL file is used in this work to define a design space, binarised data from CT or MRI
scans could easily be used instead, in order to define a volume representation of the design space, as
described by Hollister, et al. [28].

We should note that although modelling lattice structures from a uniform grid is a simple and
very widely used approach, it has certain limitations, such as high irregularity (staircase effect) of
the boundary surface associated with the discrete representation [29, 30]. In order to improve the fit
with a given design space and create a continuous and smooth surface, a so-called conformal
approach can be used [31]. For example, Robbins, et al. [32] recently proposed an FE mesh
generation algorithm, which moves mesh nodes close to the boundary of the design space before
populating the conformal grid with unit cells. Another approach would involve generating a
uniform grid within a given design space and then reducing the staircase effect using a relatively
high resolution grid near the design space boundaries. This last approach will be used in the
framework of this project.
 ACCEPTED MANUSCRIPT

When building the lattice structures by additive manufacturing, the strut orientation can have a
major impact on the mechanical properties and manufacturability of such structures [33, 34]. When
uniform unit cell strut orientations are used throughout the structure, self-supporting criteria can be
evaluated at the unit cell level. For conformal structures, each strut orientation may need to be
evaluated individually, thereby increasing the computational costs.

Geometric properties of the diamond unit cell. The project discussed in this work focuses on
the modelling of lattice structures with a specific cell type  the diamond unit cell  presented in
Figure 1-a. The figure also shows a set of eight cells (4 filled and 4 void), which are copied to form
a lattice structure. The diamond structure has been shown to be more compliant than other unit
cells, such as cubic or octahedral [35, 36]. Results by Quevedo Gonzalez and Nuno [35] also

PT
suggest that such a cell may be well suited for orthopaedic applications.

The main criterion for the unit cell selection was its compliance. While other cells, such as

RI
tetrahedron cell for example, may have significantly higher strength and stiffness, the diamond
structure has been shown to be significantly more compliant [36]. Moreover, because of the
complex internal structure created by porosity gradients, it was considered impractical to put other

SC
constraints on the unit cell geometry, such as isotropy or symmetry.

The diamond unit cell also has an advantage in the context of AM. The minimum overhang angle
NU
(as shown in Figure 1-b), which is required to consider the strut as self-supported, depends on many
factors, such as the type of AM process, the material used, and the strut length. The struts of the
diamond unit cell discussed in this work are oriented with an angle of 35.3° with respect to the build
plane, which is deemed enough to ensure their self-support during manufacturing.
MA
ED
PT
CE

Figure 1: a) Diamond lattice structure with b) tetrahedral unit cell with overhang angle
AC

Surface representation strategy. As previously mentioned, the approach selected for modelling
the lattice structure consists in populating the uniform grid with a faceted unit cell. In order to
reduce the memory requirements and keep STL files at a manageable size, the number of facets
used to model the unit cell needs to be reduced to a minimum. However, with fewer facets, it can be
difficult to eliminate sharp edges, which could lead to stress concentrations within the modelled
structure. A compromise between the number of facets used to represent the struts of the structure
and the level of stress concentration is therefore necessary. A unit cell modelled with truncated
tetrahedral nodes and four hexagonal struts is considered in this work as being the best arrangement
to simultaneously meet the requirements of the smallest number of facets and the lowest stress
concentration.

In order to create a valid STL file, the lattice structure is defined by a single tessellated shell.
While populating the uniform grid with tetrahedral unit cells, the MATLAB routine performs the
operations represented in Figure 2 to ensure that the STL file created is valid. First, the network of
hexagonal struts is modelled based on the desired cell properties. When the unit cell is being
 ACCEPTED MANUSCRIPT

modelled, each strut is represented by 12 STL facets. Then, the truncated tetrahedron nodes are used
to connect the struts together. The central node is rotated by 90° with respect to the central axis of
the cell to allow connectivity with other nodes. Finally, a verification is performed to eliminate
interior facets.

PT
RI
Figure 2: Modelling a tetrahedral unit cell with a) hexagonal strut surfaces, b) truncated
tetrahedral nodes, c) tetrahedral cell surfaces

SC
The diamond structure can be modelled by the successive repetition of the tetrahedral unit cell
and void space. The tetrahedral unit cell can, for its part, be modelled by a network of struts and
NU
nodes. Figure 3 shows a diamond lattice structure modelled using the discussed strategy, as well as
the repeated in space tetrahedral unit cell shell.
MA
ED
PT

Figure 3: Lattice structure resulting from the repetition of tetrahedral unit (node and struts)
cells and void
CE

Modelling porosity gradient. In order to model a porosity gradient, a scaling factor is applied to
each node of the lattice structure. Figure 4 shows the different steps required for such modelling in a
AC

single cell. First, scaling factors are applied to each node, and struts with varying cross-sections are
modelled to connect these nodes. Next, the truncated tetrahedrons are positioned and scaled by a
specified factor. A verification is finally performed in order to remove any interior surfaces.
 ACCEPTED MANUSCRIPT

Figure 4: Steps for modelling a graded tetrahedral unit cell a) Bare structure with defined
scaling factors b) Modelling of varying cross-section struts c) Scaled truncated tetrahedrons, and d)
Scaled tetrahedral unit cell

When modelling a lattice structure, scaling factors can be applied individually at each node. If
the lattice is composed of several thousand cells, this task can however become extremely tedious
or even impossible. In order to resolve this issue, an interpolation strategy based on kriging was
implemented [37]. Through the strategy, scaling factors can be applied for arbitrary nodes of the
grid, and then an interpolation can be performed to define factors for the remaining nodes. Note that
the implementation of the kriging strategy in the MATLAB routine is explained in section S2 of the
provided Supplementary information. The strategy can be particularly interesting when mapping the

PT
material density from bone density information. Figure 5 shows an example of a graded lattice
structure generated using this strategy. In Figure 5-a, scaling factors with defined positions and
values are represented by coloured dots. Coloured cubes are used to represent the cells of the grid

RI
which is to be populated. Figure 5-b shows the resulting lattice structure generated by applying the
scaling factors to each node of the grid using kriging interpolation.

SC
NU
MA
ED

Figure 5: Modelling of a density gradient: a) Interpolation between specified values of scaling

factors, and b) Graded lattice structure obtained from the density map
PT

Design space for biomedical applications. The tetrahedral unit cell used to populate the uniform
grid is parameterized in order to allow some control over the cell properties. Porosity, pore
interconnectivity and pore size greatly influence the mechanical and biological properties of cellular
CE

materials [12].

Arabnejad, et al. [38] described a comprehensive strategy for assessing the interplay between the
unit cell morphology, the mechanical properties of a lattice structure made of tetrahedral and
AC

octahedral cells, the bone ingrowth criteria for bone replacement materials, and the additive
manufacturing constraints. In this work, this approach for intuitive visualization of the relationship
between the pore size, strut thickness and density was adapted to the diamond structure.

In their study, Arabnejad, et al. [38] considered a minimum porosity of 50% and a pore size
varying between 50 and 800 µm as biological constraints to allow bone ingrowth. A minimum strut
thickness of 200 µm was also defined in order to consider the manufacturing constraints of AM
equipment used for bone replacement applications. In the current study, the same criteria are
considered with the exception of a minimum strut thickness, which was set to 300 µm. With the
manufacturing parameters that will be described later in this work, the fabrication of struts proved
to be unreliable below this value.

Figure 6 shows some characteristic dimensions (design variables) of the diamond lattice
structure and the design space defined by biological and manufacturing constraints. The
significance of the bold dots in the graph will be discussed in the next section. The strut thickness is
 ACCEPTED MANUSCRIPT

defined by the height of the hexahedral cross-section, while the pore diameter is defined by the
major pore size, which corresponds to the diameter of the largest sphere that can be contained
within the structure [39]. With these two parameters, the cell size (see Figure 6-b) and porosity are
determined.

PT
RI
SC
Figure 6: Characteristic dimensions of the lattice structure and design space for manufacturing
NU
and biological constraints

Finite element prediction of the stiffness and strength of the lattice structure. In order to use
the modelled lattice structures for load-bearing applications, it is essential to be able to predict their
MA

mechanical stiffness and strength. In the current work, FE modelling was performed on structures
with densities of 20, 42 and 60% [21].

A simple MATLAB routine was used to replace the STL facets of the model by triangular areas
ED

and generate an input file for ANSYS Mechanical APDL 15.0 (ANSYS, Canonsburg, PA, USA).
The ANSYS VA function was then used to create a volume from the areas. Six points from the
tensile stress-strain curve of a fully dense, additively-manufactured Ti-6Al-4V specimen were used
PT

to generate a multilinear material model for the analysis. The stiffness and yield stress evaluated for
the dense material are presented in Figure 7. Note that section S1 of the Supplementary information
provides more details on the geometry of the specimen and testing parameters. In addition to the
CE

stress-strain information, the Poisson ratio, which was not evaluated experimentally, was set to 0.3.
AC

0.2%

Figure 7: Stress-strain diagram and multilinear approximation

 ACCEPTED MANUSCRIPT

All nodes of the bottom face of the modelled structure were fixed and a vertical displacement
was imposed on those of the top face, thus generating tensile loading conditions. The apparent
stiffness of the structure was calculated by retrieving the nodal reaction forces and evaluating the
apparent stress on the bottom surface of the structures. The Young’s modulus was finally calculated
from the imposed tensile strain and the calculated apparent stress.

A convergence analysis (2% allowable change) was performed on the apparent stiffness of the
tetrahedral unit cell, using tetrahedral finite elements (SOLID187), in order to evaluate the required
finite element size. Another convergence analysis (also 2% allowable change) was performed to
determine the number of cells necessary to simulate the structures. It was found that a cubic lattice
structure defined within a grid of 11 X 11 X 11 cells was required to adequately evaluate the

PT
apparent stiffness of the structure.

Figure 8 shows the converged FE mesh for each tetrahedral unit cell as well as the 42% dense

RI
lattice structure. The number of elements necessary to simulate each of the structures is also
presented in the figure. The total displacement was applied with 10 to 20 calculation steps. The
apparent stress and strain evaluated for each of the substeps were used to draw the stress-strain

SC
diagrams, and these diagrams were then used to determine the apparent Young’s modulus and yield
stress of the structures. NU
MA
ED

Figure 8: Converged solid finite element mesh of tetrahedral unit cells of different densities (20,
42, 60%), and 11 x 11 x 11 cell grid of the 42% dense lattice structure
PT

Tensile specimen design, fabrication and testing. Tensile testing was used to experimentally
evaluate the stiffness and resistance of the lattice structures and to define scaling relations for their
CE

macro-scale modelling. While strut thickness and pore size may vary within graded structures
modelled with the proposed approach, it was decided to maintain a constant cell size. In order to
define scaling relations to predict macro-scale properties of a graded structure, three structures with
AC

a cell size of 0.83 mm and respective densities of 20, 42 and 60% were designed. The density range
was limited by the fabrication process. Structures with densities below 20% could not be
manufactured for the specified cell size, since the strut thickness was too small to be reproduced,
whereas those with densities exceeding 60% would have created issues with powder removal.

The properties of these structures are represented by bold dots on the graph of Figure 6. It is
important to note that even though the 20% and 60% dense specimens did not respect all the criteria
for biomedical applications described previously, they did respect the manufacturing criterion. For
the specified cell size, studying the structure properties outside the prescribed design space allowed
the analysis in a wider density range, thus resulting in a more accurate FE model validation.

The tensile specimens were designed with a fully dense “grip” and lattice “gauge” regions, with
the latter modelled with the MATLAB routine discussed earlier. Figure 9 shows the overall
dimensions of the specimens, and gives the geometric characteristics of the three porous regions of
varying densities. The specimens were designed to be built with the loading axis parallel to the
build direction in order to facilitate their removal from the build plate.
 ACCEPTED MANUSCRIPT

PT
Figure 9: a) Tensile specimen designs, b) fabricated specimen, and c) porous structure
properties

RI
Prior to the specimens being manufactured, STL files of the cubic porous regions were generated
using the MATLAB routine. The STL files were then transferred into MAGICS 17.02 (Materialise,

SC
Leuven, Belgium), where they were merged with the STL files of the fully dense grip sections.
Support structures were then designed to ensure a solid fixation of the specimens to the build plate.
Next, five specimens for each density were fabricated using laser powder-bed fusion technology (L-
NU
PBF) with an EOS M 280 system (EOS GmbH, Munich, Germany), EOS Ti64 powder and Ti64
Performance parameters (30 μm layer thickness). Once built, compressed air was blown through the
porous structures. A pneumatic shaker was also fixed to the build plate which was placed upside
down and left to vibrate for one hour in order to remove unmelted powder. The specimens were
MA

then heat-treated in an inert argon environment at 800°C for 6 hours, and removed from the build
plate with a band saw.

To analyze the internal structure of the as-built specimens, one specimen with a 42% dense
ED

structure was inspected using a Nikon XT H 225 (Nikon, Brighton, MI, USA) X-ray micro-
tomography system. A tube voltage of 224 kV and current of 200 µA were applied to the source.
The volume was then reconstructed in the 3D CT (Nikon, Brighton, MI, USA) software with a
PT

voxel size of 24.1 µm and analyzed in the Dragonfly V2 (Object Research Systems, Montreal, QC,
Canada) software environment.
CE

Tensile testing of the specimens to their failure was performed on an MTS Landmark system
(Eden Prairie, MN, USA), with a 0.01 mm/s crosshead speed. A layer of white paint and a black
speckle pattern were applied on the smooth surfaces of the dense grips. A digital image correlation
(DIC) system, ARAMIS 5M (GOM mbH, Braunschweig, Germany), was used to measure
AC

displacements between the top and the bottom grip sections of the specimens. The displacements
between five pairs of points on the top and bottom surfaces were measured, and the mean value was
used to calculate the apparent stress-strain diagrams for each of the specimens.

From the stress-strain diagrams, the stiffness and the yield stress of each of the specimens were
evaluated. The results obtained were first compared to those obtained via FEA, after which the
scaling relations were derived using the strategy adopted by Gibson and Ashby [22]. The strategy
consists in using a power function linking the density and the mechanical properties such as
stiffness and yield stress. Equations (1) and (2) show the power functions defining the apparent
stiffness and yield stress of a porous structure, where E * and y* represent the apparent stiffness and
yield stress,Es , and y, the stiffness and yield stress of the bulk material,  , the relative density of
the material, and C1 , C2 , n1 and n2 are the coefficients which must be determined for a given
porous structure.
 ACCEPTED MANUSCRIPT

E*  EsC1 n1 (1)

 *y   ysC2  n2
(2)

RESULTS

Mechanical properties evaluated by finite element simulations and tensile testing. The
calculated (FEA) stress-strain diagrams of the lattice structures are presented in Figure 10-a. The
apparent stiffness was calculated by applying the Excel slope function to apparent stress and strain
data on the linear parts of the curves. The apparent yield stress was then determined at a 0.2% strain

PT
offset. A linear stress-strain dependence followed by structural softening is observed in all the
cases. The mechanical behavior of the simulated lattice structures is therefore similar to that of fully
dense ductile materials.

RI
SC
NU
MA
ED

Figure 10: Stress-strain diagrams of 20, 42 and 60% dense lattice structure from a) FEA results
and b) tensile testing results

The mean Young’s modulus and yield stress evaluated from the experimental stress-strain
PT

diagrams are presented in Figure 10-b. These results show very little stiffness and yield stress
variability among the specimens of the same density. For the 42% dense specimens, the maximum
standard deviations for the Young’s modulus and yield stress values were evaluated as being
CE

0.15 GPa and 3.41 MPa, respectively.

Comparison of the FEA and experimental results. The stiffness and yield stress evaluated by
AC

FEA are compared to the experimental results in order to determine if the simulation strategy allows
an accurate evaluation of the mechanical properties. Table 1 shows that there are notable
discrepancies between the FEA and the experimentally-measured data, where the maximum
deviations for the Young’s modulus and yield stress values corresponded to 40% and 50%,
respectively.

Table 1: Comparison of experimental and FEA results for Young’s modulus

Young's modulus (GPa) Yield stress (MPa)

Variation, Variation,
Density FEA EXP. FEA EXP.
% %
20 2.7 1.6 +40 45 23 +50
42 12.6 7.6 +40 165 90 +45
60 26.8 20.3 +24 265 181 +32
 ACCEPTED MANUSCRIPT

Scaling relations. Figure 11 shows the apparent Young’s modulus (E) and yield stress (σy) of
the porous structure specimens as functions of density. The coefficients of the power laws used for
the scaling relations were determined by applying the least squares method. The values for
coefficients C1 , n1 , C2 and n2 determined from the modelling, and experiments are provided in
Table 2.

PT
RI
SC
NU
Figure 11: Scaling relations for the modelled and experimentally-measured Young’s modulus
MA

and yield stress values

Table 2: Scaling relations coefficients obtained from FEA and experiments
ED

C n
E* σy* E* σy*
FEA 0.67 0.59 2.07 1.72
PT

EXP. 0.52 0.41 2.25 1.87

With Es = 114 GPa and σys = 1120 MPa for fully dense
material
CE

DISCUSSION
AC

For the three tested densities (20, 42 and 60%), the stiffness of the corresponding lattice
structures lies within the stiffness range of the cortical and trabecular bones (0.3 to 30 GPa [40,
41]), whereas the yield stresses fall within the same bone values (from 84.9 to 107.9 GPa,
Bayraktar, et al. [42]). Among these three structures, the 42% dense structure most closely matches
the pore size- and density-related criteria for bone scaffolds, in addition to respecting the AM
manufacturability criteria. In future work, the scaling relations and design space developed and
partially validated in this work can be used to create new lattice structures with adjustable
properties.

Significant discrepancies were found in the current work between the simulations and
experiments. These discrepancies may be due to the geometric deviations between the modelled and
fabricated structures. Bagheri, et al. [33] and Gonzalez and Nuno [34], demonstrated that the
geometric deviations of lattice structures produced by additive manufacturing can have a substantial
influence on the mechanical properties of the structures. Irregularities generally have the effect of
reducing mechanical properties such as stiffness and strength, which may explain why FEA results
 ACCEPTED MANUSCRIPT

for these properties are consistently higher than experimental results. Figure 12 shows an X-ray
image of the 42% dense specimen, along with the reconstruction of a part of the porous region. In
Figure 12-b, the surface outline of the reconstructed volume is also presented. The volume of the
reconstructed porous region was highly dependent on the threshold value used to distinguish
material from void, and we were not able to distinguish unmelted powder from dense material. It
was therefore not possible to validate that all such powder had been removed from the porous
structure.

A greyscale range corresponding to the material/void threshold was set manually and resulted in
a variation between 38% and 45% of the density for the reconstructed porous region. Unfortunately,
the volume reconstruction of the 42% dense specimen showed that the actual density of the

PT
fabricated specimen could diverge by 4% from the target. Therefore, the analysis did not allow
any specific conclusions to be made regarding the conformity of the fabricated part. It was therefore
not possible to confirm or infirm the geometric deviations-based hypothesis on the main cause of

RI
discrepancies between the FEA and experimental results observed in this study.

SC
NU
MA
ED

Figure 12: Volume reconstruction of a 42% dense specimen. a) X-ray image of the specimen. b)
Slice of the reconstructed volume
PT

The volume reconstruction of one 42% dense specimen was also used to examine its internal
structure. Upon visual inspection, all the struts appeared to have been correctly fabricated. Slight
surface irregularities were however visible. The geometric deviations with respect to the CAD
CE

model due to these irregularities were not evaluated. Further investigations will be necessary to
evaluate whether the irregularities observed in the current work could explain the discrepancies in
FEA and experimental results.
AC

An average mass of 73.17 g was measured for the five 42% dense manufactured specimens,
which was 5% smaller than their theoretical mass calculated from the CAD volume and material
density, with the latter taken from the EOS Ti64 datasheet. This difference in mass provides an
additional indication of geometric deviations. A reduced mass may indicate a generally smaller strut
thickness and contribute to decreasing the bending stiffness and strength of some individual struts.
On the macro scale, this would cause a reduction in apparent stiffness and yield stress. This may
therefore explain why experimental results were consistently lower than those obtained via
modeling.

Figure 13-a shows one of the five fractured specimens for each of the designed lattice structures.
For the 42% and 60% dense specimens, fractures appeared to occur by cleavage, along a plane
going through the struts of the structures. During tensile testing, the shear stress in the plane was
maximal, as the unit cell axis coincided with the principal stress directions. This fracture plane,
which coincides with the <111> direction shown in Figure 13-b, is the plane of minimum cohesion
and dominant cleavage in physical diamonds [43].
 ACCEPTED MANUSCRIPT

PT
Figure 13: a) Fracture of specimens for the three tested densities. b) Minimum cohesion plane of
physical diamonds

RI
It should be noted that while fractures occurred along a predominant plane for the 42 and 60%
dense structures, three of the five 20% dense specimens were severed along a horizontal plane at the

SC
interface between the porous region and the dense grips. The stress concentration due to a high
stiffness mismatch between the dense and porous parts of the tensile specimens may explain why
fractures initiated and propagated at the interface. In similar circumstances, to reduce such a stress
concentration, Simoneau, et al. [23] increased the porosity of their tensile specimens from the grip
NU
to the gauge section gradually. In our next work, the porosity gradient modelling feature discussed
earlier in this work will be used to generate such a gradient, and to avoid the near-the-grip fracture
of tensile specimens.
MA

The stress and strain data used for the bulk material model may also have contributed to
variations between the FEA and experimental results. The data were taken from an additively
manufactured tensile specimen oriented at 45° with respect to the build plate. The stiffness of the
ED

tensile specimens oriented at 0°, 45° and 90° were evaluated at 107, 115 and 108 GPa respectively.
The data from the 45° specimen were used for the FEA, as that orientation was the closest to the
lattice strut orientation (35.3°). However, the difference in stiffness observed may have been due to
process variations rather than specimen orientation. Reducing the stiffness in the bulk material
PT

model would have lowered the stiffness of the lattice structures and reduced the gap between the
FEA and experimental results. In future work, a sensitivity analysis may be required in order to
determine the effects of bulk material properties on the apparent properties of the lattice structures.
CE

CONCLUSIONS
AC

In this work, a novel approach was proposed for the geometric modelling of porosity graded
lattice structures within the MATLAB environment. The approach serves as an alternative to CAD-
based modelling, which can require much more computational power. This approach can be used to
efficiently generate diamond-type lattice structures with properties well suited for biomedical
applications. Other considerations, such as strut thickness, were also discussed in order to ensure
that structures can be fabricated via the powder bed fusion technology.

Tensile testing and FE analysis were performed in order to determine the stiffness and yield
stress of the modelled structures. The main objective of this characterization was to validate the
numerical model of the lattice structures, thus creating an adequate numerical tool for FE
simulations and structural optimization of porous load-bearing implants. A comparison of the
results obtained by the two methods showed notable differences. Geometric irregularities of the
manufactured structures were considered as the main causes of these discrepancies, but this
hypothesis could not be validated due to the insufficient accuracy of the specimen reconstruction. In
future works, a better evaluation of the CAD model – AM-built structure conformity could
highlight the real cause (causes) for these deviations and help in developing compensation strategies
 ACCEPTED MANUSCRIPT

for the improved FE simulations. The data used for the multilinear material modelling may also
have contributed to these deviations: this hypothesis will be validated in future work. Additional
testing of the bulk material may be necessary to improve the accuracy of simulations.

From the experimental results, the coefficients of scaling relations were determined for a
specified cell size. These relations can now be used in approximating the apparent stiffness and
yield stress of structures with any desired density. Work is currently underway to use these
modelling tools and the information gathered from the mechanical testing to design a porous
femoral stem with improved compatibility with bone. The apparent mechanical properties evaluated
in the current work will then be used for the FE modelling of the porous region of the stem. The
stem for its part will be produced by additive manufacturing and tested following the procedure

PT
described by Simoneau, et al. [9].

RI
ACKNOWLEDGMENTS

SC
This work was financially supported by the Natural Science and Engineering Council of Canada
(NSERC) through the Discovery grant of professor Terriault (RGPIN-2014-05178).
NU
REFERENCES

[1] L. Murr, S. Gaytan, F. Medina, H. Lopez, E. Martinez, B. Machado, D. Hernandez, L. Martinez,

MA

M. Lopez, R. Wicker, Next-generation biomedical implants using additive manufacturing of

complex, cellular and functional mesh arrays, Philosophical Transactions of the Royal Society A:
Mathematical, Physical and Engineering Sciences 368(1917) (2010) 1999-2032.
ED

[2] A. Sola, D. Bellucci, V. Cannillo, Functionally graded materials for orthopedic applications – an
update on design and manufacturing, Biotechnology Advances 34(5) (2016) 504-531.

[3] D.C. Dunand, Processing of titanium foams,, Adv Eng Mater, 6 (2004) 369-376.
PT

[4] R. Singh, P.D. Lee, R.J. Dashwood, T.C. Lindley, Titanium foams for biomedical applications:
a review, Materials Technology 25(3-4) (2010) 127-136.
CE

[5] M.M. Shbeh, R. Goodall, Open Celled Porous Titanium, Advanced Engineering Materials, DOI:
10.1002/adem.201600664 (2017).
AC

[6] K. Hazlehurst, C.J. Wang, M. Stanford, Evaluation of the stiffness characteristics of square pore
CoCrMo cellular structures manufactured using laser melting technology for potential orthopaedic
applications, Mater Design 51 (2013) 949-955.

[7] S. Arabnejad, B. Johnston, M. Tanzer, D. Pasini, Fully porous 3D printed titanium femoral stem
to reduce stress-shielding following total hip arthroplasty, Journal of Orthopaedic Research (2016).

[8] O.L. Harrysson, O. Cansizoglu, D.J. Marcellin-Little, D.R. Cormier, H.A. West, Direct metal
fabrication of titanium implants with tailored materials and mechanical properties using electron
beam melting technology, Materials Science and Engineering: C 28(3) (2008) 366-373.

[9] C. Simoneau, P. Terriault, B. Jetté, M. Dumas, V. Brailovski, Development of a porous metallic

femoral stem: Design, manufacturing, simulation and mechanical testing, Mater Design (2016).
 ACCEPTED MANUSCRIPT

[10] S. Limmahakhun, A. Oloyede, K. Sitthiseripratip, Y. Xiao, C. Yan, Stiffness and strength

tailoring of cobalt chromium graded cellular structures for stress-shielding reduction, Mater Design
(2016).

[11] A. Maiti, W. Small, J. Lewicki, T. Weisgraber, E. Duoss, S. Chinn, M. Pearson, C. Spadaccini,

R. Maxwell, T. Wilson, 3D printed cellular solid outperforms traditional stochastic foam in long-
term mechanical response, Scientific reports 6 (2016).

[12] X. Wang, S. Xu, S. Zhou, W. Xu, M. Leary, P. Choong, M. Qian, M. Brandt, Y.M. Xie,
Topological design and additive manufacturing of porous metals for bone scaffolds and orthopaedic
implants: A review, Biomaterials 83 (2016) 127-141.

PT
[13] S. Giannitelli, D. Accoto, M. Trombetta, A. Rainer, Current trends in the design of scaffolds
for computer-aided tissue engineering, Acta biomaterialia 10(2) (2014) 580-594.

RI
[14] G. Papazetis, G.-C. Vosniakos, Direct porous structure generation of tissue engineering
scaffolds for layer-based additive manufacturing, The International Journal of Advanced

SC
Manufacturing Technology (2015) 1-13.

[15] A. Medeiros e Sá, V.M. Mello, K. Rodriguez Echavarria, D. Covill, Adaptive voids, The
NU
Visual Computer 31(6-8) (2015) 799-808.

[16] S.J. Hollister, Porous scaffold design for tissue engineering, Nature materials 4(7) (2005) 518-
524.
MA

[17] O. Fryazinov, T. Vilbrandt, A. Pasko, Multi-scale space-variant FRep cellular structures,

Computer-Aided Design 45(1) (2013) 26-34.
ED

[18] D.-J. Yoo, Recent trends and challenges in computer-aided design of additive manufacturing-
based biomimetic scaffolds and bioartificial organs, Int. J. Precis. Eng. Manuf. 15(10) (2014) 2205-
2217.
PT

[19] M. Mmillan, M. Jurg, M. Leary, M. Brandt, Programmatic Lattice Generation for Additive
Manufacture, Procedia Technology 20 (2015) 178-184.
CE

[20] A. Armillotta, R. Pelzer, Modeling of porous structures for rapid prototyping of tissue
engineering scaffolds, The International Journal of Advanced Manufacturing Technology 39(5-6)
(2008) 501-511.
AC

[21] V. Brailovski, P. Terriault, Metallic Porous Materials for Orthopedic Implants: Functional
Requirements, Manufacture, Characterization, and Modeling, Reference Module in Materials
Science and Materials Engineering, Elsevier (2016).

[22] L.J. Gibson, M.F. Ashby, Cellular solids: structure and properties, Cambridge university press
(1999).

[23] C. Simoneau, V. Brailovski, P. Terriault, Design, manufacture and tensile properties of

stochastic porous metallic structures, Mech Mater 94 (2016) 26-37.

[24] W. van Grunsven, E. Hernandez-Nava, G.C. Reilly, R. Goodall, Fabrication and mechanical
characterisation of titanium lattices with graded porosity, Metals 4(3) (2014) 401-409.
 ACCEPTED MANUSCRIPT

[25] S.A. Khanoki, D. Pasini, Multiscale design and multiobjective optimization of orthopedic hip
implants with functionally graded cellular material, Journal of biomechanical engineering 134(3)
(2012) 031004.

[26] S. Xu, J. Shen, S. Zhou, X. Huang, Y.M. Xie, Design of lattice structures with controlled
anisotropy, Mater Design 93 (2016) 443-447.

[27] J.E. Cadman, S. Zhou, Y. Chen, Q. Li, On design of multi-functional microstructural materials,
Journal of Materials Science 48(1) (2013) 51-66.

[28] S.J. Hollister, C.Y. Lin, E. Saito, C.Y. Lin, R.D. Schek, J.M. Taboas, J.M. Williams, B. Partee,

PT
C.L. Flanagan, A. Diggs, E.N. Wilke, G.H.V. Lenthe, R. Müller, T. Wirtz, S. Das, S.E. Feinberg,
P.H. Krebsbach, Engineering craniofacial scaffolds, Orthodontics & Craniofacial Research 8(3)
(2005) 162-173.

RI
[29] S. Vongbunyong, S. Kara, Rapid generation of uniform cellular structure by using
prefabricated unit cells, International Journal of Computer Integrated Manufacturing (2016) 1-13.

SC
[30] A.O. Aremu, J.P.J. Brennan-Craddock, A. Panesar, I.A. Ashcroft, R.J.M. Hague, R.D.
Wildman, C. Tuck, A voxel-based method of constructing and skinning conformal and functionally
NU
graded lattice structures suitable for additive manufacturing, Additive Manufacturing 13 (2017) 1-
13.

[31] J. Nguyen, S.-i. Park, D. Rosen, Heuristic optimization method for cellular structure design of
MA

light weight components, Int. J. Precis. Eng. Manuf. 14(6) (2013) 1071-1078.

[32] J. Robbins, S.J. Owen, B.W. Clark, T.E. Voth, An efficient and scalable approach for
generating topologically optimized cellular structures for additive manufacturing, Additive
ED

Manufacturing (2016).

[33] Z.S. Bagheri, D. Melancon, L. Liu, R.B. Johnston, D. Pasini, Compensation strategy to reduce
geometry and mechanics mismatches in porous biomaterials built with Selective Laser Melting,
PT

Journal of the Mechanical Behavior of Biomedical Materials (2016).

[34] F.J.Q. Gonzalez, N. Nuno, Finite element modeling of manufacturing irregularities of porous
CE

materials, Biomaterials and Biomechanics in Bioengineering 3(1) (2016) 1-14.

[35] F.J. Quevedo Gonzalez, N. Nuno, Finite element modelling approaches for well-ordered
AC

porous metallic materials for orthopaedic applications: cost effectiveness and geometrical
considerations, Comput Methods Biomech Biomed Engin 19(8) (2016) 845-54.

[36] A. Zadpoor, R. Hedayati, Analytical relationships for prediction of the mechanical properties
of additively manufactured porous biomaterials, Journal of Biomedical Materials Research Part A
(2016).

[37] F. Trochu, A contouring program based on dual kriging interpolation, Engineering with
Computers 9(3) (1993) 160-177.

[38] S. Arabnejad, R. Burnett Johnston, J.A. Pura, B. Singh, M. Tanzer, D. Pasini, High-strength
porous biomaterials for bone replacement: A strategy to assess the interplay between cell
morphology, mechanical properties, bone ingrowth and manufacturing constraints, Acta
Biomaterialia 30 (2016) 345-356.
 ACCEPTED MANUSCRIPT

[39] R.C.S. L. Mullen, P. Fox, E. Jones, C. Ngo, C.J. Sutcliffe, Selective laser melting: a unit cell
approach for the manufacture of porous, titanium, bone in-growth constructs, suitable for
orthopedic applications. II. Randomized structures, J Biomed Mater Res Part B. 92B (2010) 178-
188.

[40] J.D. Currey, The structure and mechanics of bone, Journal of Materials Science 47 (2012) 41-
54.

[41] E.F. Morgan, T.M. Keaveny, Dependence of yield strain of human trabecular bone on
anatomic site, Journal of Biomechanics 34(5) (2001) 569-577.

PT
[42] H.H. Bayraktar, E.F. Morgan, G.L. Niebur, G.E. Morris, E.K. Wong, T.M. Keaveny,
Comparison of the elastic and yield properties of human femoral trabecular and cortical bone tissue,
Journal of Biomechanics 37(1) (2004) 27-35.

RI
[43] R. Telling, C. Pickard, M. Payne, J. Field, Theoretical strength and cleavage of diamond,
Physical Review Letters 84(22) (2000) 5160.

SC
NU
MA
ED
PT
CE
AC
 ACCEPTED MANUSCRIPT

Graphical abstract

PT
RI
SC
NU
MA
ED
PT
CE
AC
 ACCEPTED MANUSCRIPT

Highlights

 A novel approach for the design of porosity-graded lattice structures suitable for bone
replacement applications is proposed.
 Finite element analysis is employed to evaluate the stiffness and yield strength of 20, 42 and
60% lattice structures.
 Laser powder-bed fusion technology is used to manufacture Ti-6Al-4V tensile specimens for
the experimental evaluation of stiffness and yield strength.
 The experimentally evaluated Young’s modulus and yield stress are shown to be comparable
to those for cortical and trabecular bone.

PT
 Scaling relations for stiffness and yield stress were developed and allow the computer-
assisted design of porosity-graded load-bearing implants.

RI
SC
NU
MA
ED
PT
CE
AC