Accepted Manuscript

Design for 4D printing: A voxel-based modeling and simulation

of smart materials

Germain Sossou, Frédéric Demoly, Hadrien Belkebir, H. Jerry Qi,

Samuel Gomes, Ghislain Montavon

PII: S0264-1275(19)30235-7
DOI: https://doi.org/10.1016/j.matdes.2019.107798
Article Number: 107798
Reference: JMADE 107798
To appear in: Materials & Design
Received date: 20 February 2019
Revised date: 4 April 2019
Accepted date: 14 April 2019

Please cite this article as: G. Sossou, F. Demoly, H. Belkebir, et al., Design for 4D
printing: A voxel-based modeling and simulation of smart materials, Materials & Design,
https://doi.org/10.1016/j.matdes.2019.107798

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

Design for 4D printing: A voxel-based modeling and simulation of smart

materials
Germain Sossou1, Frédéric Demoly1, Hadrien Belkebir1, H. Jerry Qi2, Samuel Gomes1, Ghislain
Montavon1

1
ICB CNRS 6303, Univ. Bourgogne Franche-Comté, UTBM, 90010 Belfort, France
2
G.W.W. School of Mechanical Engineering, Georgia Institute of Technology, Atlanta, GA 30332, USA

T
IP
Abstract

CR
In the wake of the breakthrough being by additive manufacturing (AM), there is another rapidly growing
manufacturing progress: 4D printing (4DP). It is basically AM with, inter alia, smart materials (SMs).
Owing to the stimulus-responsive behaviors of these materials, the parts so manufactured are imbued with
the ability to change. 4DP is being given huge research efforts regarding its manufacturing aspects.

US
However so little is made to let designers explore the so uncovered design space. Compared to DFAM,
what could be called Design for 4D printing (DF4DP) does lag far behind. In this article, a modeling
framework for simulating SMs and conventional materials behaviors on a voxel basis is proposed; this
AN
allows for arranging materials in any distribution and rapidly evaluating the behavior of the distribution.
Homogeneous and heterogeneous objects made of conventional materials and SMs were modeled and
simulated. The modeled SMs were limited to non-programmable shape changing SMs including:
M

piezoelectric material, electro-/magneto-/photostrictive materials and hydrogel. A printed smart valve and
a theoretical actuator (both from other publications) were used as test cases. These simulations have a
speed reasonable for the design iterations needed in conceptual design phase and they yield results in good
ED

agreement with physics.

PT

1. Introduction
Since its discovery in 1987 (under the stereolithography apparatus patent [1]) additive manufacturing has
CE

evolved from a prototyping process to a fully established manufacturing process. Mainly praised is its
shape complexity characteristic, indeed thanks to this capability shapes that are infeasible with
conventional (subtractive) manufacturing processes are now manufacturable. In addition to shape
AC

complexity, other characteristics are among the 3D printing revolution engines: hierarchical complexity
(features size at almost any length scale can be realized within the same part), functional complexity
(mechanisms – with sometimes embedded electronics – can be manufactured without any assembly
operations), and material complexity (parts with any material distribution – MD – and properties are now
feasible). This latter capability – such as reviewed in [2] – is best illustrated with the PolyJet [3] AM
technique. This technique works by selectively depositing tiny droplets of UV curable resins, smoothing
them in a thin layer and curing the layer, and by repeating this process in a layer-by-layer manner. Up to 3
base resins can be mixed into any ratio to generate materials with a large range of properties including
color, transparency, shore hardness, and many others. The obtained mixtures are the so-called digital
materials. Parts with almost any MD (and thus multiple properties) can then be printed. Fig.1 (a) shows an
illustration of a multi-material part made by the PolyJet technique. The material complexity allowed by
AM has been further demonstrated by Katsumi et al. [4] who developed a 3D printing machine for

1
 ACCEPTED MANUSCRIPT

depositing metal. Their machine has been used to print a functionally graded material whose properties
range from a metal (with low meting temperature) to a polymer as shown in Fig.1 (b). As shown in [2]
others AM techniques such as FDM or even SLA have been demonstrated to be able to print multi-
material objects.

T
IP
CR
US
(a) (b)

Fig.1 - Additive manufacturing's material complexity demonstrated by (a) a PolyJet printed model of human
head [3], and (b) a functionally graded material made of metal and polymer [4]
AN
In addition to the widely praised shape complexity allowed by AM, material complexity is expanding
further the design space now available to designers, making creativity and imagination the main barriers.
When the kinds of materials involved are taken into consideration the design freedom is more enlarged.
M

Furthermore, the design space allowed by AM is being expanded further thanks to the interaction AM –
smart materials (SMs), which has been coined as 4D Printing [5]. Here SMs are materials whose state
ED

changes upon exposure to a specific stimulus. These materials owe their smartness to both what they are
sensitive to and how they respond to the stimuli. Examples include thermochromic materials [6], which
change color under heat, magnetostrictive [7] materials which deform upon exposure to a magnetic field
PT

or electrorheological fluid [8] whose viscosity change with electricity. The characterization of their
smartness can also be extended to whether their behaviors are reversible or not. The aforementioned
material complexity of AM has been a main catalyst of what can be called the “4D revolution” making the
CE

AM-SMs interaction an attractive research topic. Indeed despite its infancy, many review papers have
already been published on the topic [9-12]. The vast majority of current research work of 4DP is targeted
at its manufacturing aspects (e.g. development of new materials [13], demonstration of new manufacturing
routes for 4DP [14], etc.). However so little is made to make the so gained knowledge on 4DP available in
AC

a form usable by designers. In other words, compared to Design for Additive Manufacturing [15], what
could be called Design for 4D printing (DF4DP) does lag far behind. A few works [16-19] have already
attempted to fill that gap regarding shape memory polymers (SMP). These consist in constitutive
modeling and development of bespoke finite element models for simulating the thermomechanical
training of SMP and the subsequent shape recovery. What is sought through the use of SMs as raw
materials in AM is basically to imbue structures with a smart behavior, in such a way that the material
is/becomes the mechanism [20], and a passive source of energy (available in the environment or supplied
internally) is what moves the mechanism to produce the desired/designed behavior. From a designer’s
point of view, questions that may then arise include: can one single SM be sufficient to produce a desired
behavior? If no, what other materials should be combined to it? How can SMs be ‘mixed’ to produce a
behavior? In a given spatial arrangement, how would a SM behave once subjected to the stimulus? Etc.

2
 ACCEPTED MANUSCRIPT

Answers to these questions require the capability to model and simulate the behaviors of SMs, especially
in the conceptual design phase of the design process. It can be seen in the SMs-based actuators presented
in [21] that the way SMs are spatially combined to another material in a part is key to how the part
behaves upon exposure to the stimulus. Taken alone a SM would simply exhibit its basic behavior
(whatever the size of the part made of it), however combined with a conventional material the resulting
behavior may not be intuitive.
When designing for 4DP, there is therefore, inter alia, the need at some point to model and simulate the
behavior of the considered SMs. Such modeling scheme should be at a granularity that allows for a
seamless “intertwining” of SMs with other materials. Finally as designing the right distribution of SMs to

T
achieve a desired functionality is not an intuitive task, the modeling scheme should also make design
iteration easy and fast.

IP
The objective of our work is to develop a voxel-based modeling framework that is intended to be used in

CR
the conceptual design phase. This voxel-based modeling framework is targeted at rapidly knowing how a
MD would behave. As such the focus is on rapidly getting a qualitative answer to how a given distribution
of SMs would behave. In this paper, which is the 1st part of a series of two, the framework for modeling
and rapidly simulating the behavior of SMs is presented. In section 2, the foundations of our modeling

US
scheme are established. Building on this, section 3 shows how SMs are modeled; a few simulations are
run to validate the proposed framework in section 4. Finally conclusions are drawn and future work is
stressed out.
AN
2. Voxel mechanics modeling setup
M

As explained in the introduction section, regarding SMs, one MD generally equates to one concept. The
explorative 3D painting approach to determine a distribution – that is, the ability given to designers to
ED

pattern materials in any arrangement in order to test their distributions’ behaviors – make it paramount to
work on a framework which is fast (not too computationally costly). Moreover, this need for rapid
iteration should not be at the cost of accuracy, even though the focus is mainly on a qualitative answer of
PT

how a given distribution would behave or what distribution would yield a prescribed behavior. For such
modeling framework to be as physically realistic as possible and easily usable, it is required to:
CE

- Reproduce actual behavior of matter – i.e. any deformation must be physically plausible and
accurate.
- Be volumetric, as shape is involved – The actual three-dimensional shape of any modeled
AC

distribution should be reproduced.

- Be sensitive – Its behavior should be driven by changes in the environment. It must depend on
variable measuring how much stimulus is sensed, so that behaviors are triggered accordingly.
- Capture the actual behavior of the modeled SM – Again, while the focus is not on accuracy, the
proposed models for SMs should be based on declarative knowledge such as their existing
constitutive equations.

3
 ACCEPTED MANUSCRIPT

Material
2 Discretization 3
modeling
Deformation
4 computation

T
IP
Computation of

CR
Rough design
1 5 the voxelized
space solution
object deformation

Fig. 2 – Continuum mechanics modeling setup

US
As shown in Fig. 2 the proposed approach for modeling conventional material behavior is organized in
five main steps: (1) definition of rough design space, (2) discretization, (3) material modeling, (4)
deformation computation and (5) computation of the voxelized object deformation. These steps are
AN
described in the following subsections. Such approach assumes that material properties in a single voxel
are homogeneous, i.e. a voxel is made of a single homogeneous material.
M

The proposed methodology for computing the voxelized object deformation takes advantage of a well-
established technique used in the computer graphics (CG) area: skinning; all the CG techniques bringing
3D characters to life fall under the umbrella of this term. Skinning [22] is the process of controlling
ED

deformations of a given object using a set of deformation primitives, which are transformations associated
with bones of an animation skeleton. Roughly speaking, the overall skinning process can be described as:
(1) skeleton extraction from the character geometry, the skeleton is then embedded within the geometry. A
PT

skeleton is composed of bones and joints. (2) Each vertex of the character is assigned a set of weights
(each corresponding to a bone) quantizing how much that vertex is affected by transformations applied to
the bones. (3) Applications of transformation (which are relative rotations) to the bones to animate the
CE

character.

2.1 Design space definition and discretization (Steps 1 and 2 of Fig. 2)

AC

Here an approach similar to mass-spring modeling has been considered to simulate the mechanical
behavior of matter. First of all, the geometry to be simulated is defined as the design space (step 1). Then
the so defined shape is discretized – regardless of its (possible or intended) MD – into equally sized cubes:
the voxels. Such representation and discretization of matter, despite being – depending of the chosen voxel
size – at the cost of the accuracy of the represented geometry, allow for a finer control over the MD. In
addition, chunking a shape into voxels does make the process of specifying a MD more intuitive compared
to the use of spatial field functions [23] or others explicitly defined material functions [24]. As a shape is
physically involved, spatial reasoning helps the cognitive aspect of the design process.
Voxels are connected (from their centers) not by springs but by three-dimensional beams. These beams
form a 3D lattice frame, which acts as a backbone (or a control structure) of the whole shape. As they are

4
 ACCEPTED MANUSCRIPT

extending from center to center, the beams are then initially of a voxel size. The frame is what “holds” the
matter together and that governs the deformation of the whole geometry.

2.2 Material modeling (Step 3 of Fig. 2)

Each voxel is assumed to be homogeneous and made of a single linear and isotropic material. Thus, the
material properties required to fully characterize its mechanical behavior are limited to Young modulus
(E) and shear modulus (G). As explained in section 3, for SMs, other material properties are added to

T
account for their stimulus-responsive behavior.

IP
As it is the frame deformation, which drives the object deformation, voxels’ materials properties should be
mapped to the beam material properties. To account for this, we have used an inheritance scheme: beams
materials properties are inherited from the pairs of voxels they are connecting. In case where two voxels

CR
are made of the same material, the beam material properties are the same but when two voxels are of
dissimilar materials, composite values for the material properties are taken for them, as described by the
formulae:

US
2𝐸1 𝐸2 2𝐺1 𝐺2
𝐸𝑐 = , and 𝐺𝑐 = (1)
𝐸1 + 𝐸2 𝐺1 + 𝐺2
AN
2.3 Deformation computation (Step 4 and Step 5 of Fig. 2)
Computation of the skeleton frame deformation (Step 4 of Fig. 2)
M

The beams forming the whole frame are modelled as 3D Euler-Bernoulli beams which resist axial,
bending and twisting actions. Let l denote, the voxel size. The beam cross-section properties, assumed to
ED

be along the X axis, are as follows:

- Beam moments of inertia about the neutral (x) axis:
PT

𝑏 ∙ ℎ3 𝑙 4
𝐼𝑦𝑦 = 𝐼𝑧𝑧 = 𝐼 = = (2)
12 12
CE

- Beam moment of inertia characterizing torsional rigidity:

𝑏ℎ(𝑏 2 + ℎ2 ) 𝑙 4
𝐽𝑥𝑥 = 𝐽 = = (3)
AC

12 6

- Cross section area: 𝐴 = 𝑙 2

The direct stiffness method [25] is used to compute the frame’s degrees of freedoms (DOFs) and
subsequently its deformed shape. Each node (again, which is a voxel’s center) has six DOFs and the
stiffness matrix for a beam oriented into the positive x direction is expressed (in its local coordinate
system) as follows:

5
 ACCEPTED MANUSCRIPT

𝑅𝑎 0 0 0 0 0 −𝑅𝑎 0 0 0 0 0
0 𝑅𝑧3 0 0 0 𝑅𝑧2 0 −𝑅𝑧3 0 0 0 𝑅𝑧2
0 0 𝑅𝑦3 0 −𝑅𝑦2 0 0 0 −𝑅𝑦3 0 −𝑅𝑦2 0
0 0 0 𝑅𝑥 0 0 0 0 0 −𝑅𝑥 0 0
0 0 −𝑅𝑦2 0 2 ∙ 𝑅𝑦 0 0 0 𝑅𝑦2 0 𝑅𝑦 0
̅̅̅ 0 𝑅𝑧2 0 0 0 2 ∙ 𝑅𝑧 0 −𝑅𝑧2 0 0 0 𝑅𝑧
𝐾𝑒 = (4)
−𝑅𝑎 0 0 0 0 0 𝑅𝑎 0 0 0 0 0
0 −𝑅𝑧3 0 0 0 −𝑅𝑧2 0 𝑅𝑧3 0 0 0 −𝑅𝑧2
0 0 −𝑅𝑦3 0 𝑅𝑦2 0 0 0 𝑅𝑦3 0 𝑅𝑦2 0
0 0 0 −𝑅𝑥 0 0 0 0 0 𝑅𝑥 0 0

T
0 0 −𝑅𝑦2 0 𝑅𝑦 0 0 0 𝑅𝑦2 0 2 ∙ 𝑅𝑦 0
( 0 𝑅𝑧2 0 0 0 𝑅𝑧 0 −𝑅𝑧2 0 0 0 2 ∙ 𝑅𝑧 )

IP
Where:
𝐸𝐴 𝐺𝐽

CR
𝑅𝑎 = , 𝑅𝑥 =
𝐿 𝐿
2𝐸𝐼𝑦𝑦 6𝐸𝐼𝑦𝑦 12𝐸𝐼𝑦𝑦
𝑅𝑦 = , 𝑅𝑦2 = 2
, 𝑅𝑦3 = (5)
𝐿 𝐿 𝐿3

US
2𝐸𝐼𝑧𝑧 6𝐸𝐼𝑧𝑧 12𝐸𝐼𝑧𝑧
{ 𝑅𝑧 = 𝐿 , 𝑅𝑧2 = 2 ,
𝐿
𝑅𝑧3 =
𝐿3
AN
As the beams are located on a regular grid, each of them can be assumed to be oriented either into the
positive x direction, y direction or z direction. The transformation matrix that is used to express their
stiffness matrix in the global coordinate system (GCS) can therefore be easily pre-computed depending on
M

their direction.
The nodes’ DOFs, expressed in GCS, are represented by a vector:
ED

𝑈𝑖 = [𝑥𝑖 𝑦𝑖 𝑧𝑖 𝜃𝑥𝑖 𝜃𝑦𝑖 𝜃𝑧𝑖 ] (6)

The whole system of equations is then formulated as:
PT

𝐾𝑈 = 𝐹 (7)

Where K is the global stiffness matrix, 𝑈 = [𝑈𝑖 … 𝑈𝑛 ]𝑇 (n: number of voxels) and F a 6n-vector
CE

containing the boundary conditions (including forces and moments and prescribed displacements and
rotations). Once the DOFs are computed, they are used to compute the beam-deformed shape and
ultimately the object deformed shape, as described in the following paragraph.
AC

Computation of the voxelized object deformation (Step 5 of Fig. 2)

The voxelized object’s deformation is controlled by the deformation of the beams. The idea of using a 3D
lattice frame as a control structure of the whole shape is somewhat inspired by skeleton driven animation
or more generally skinning, as introduced at the beginning of the section. The technique has been used as
it is used by graphic designers (especially those involved in animation) except that the motion of the
skeleton is physics-based (instead of user defined). Harnessing the relative simplicity of skinning to our
voxel-based modeling and simulation approach is also a way of bridging the computational graphics, the
computational mechanics and the manufacturing communities. Communities which may all have interests
in 4DP.

6
 ACCEPTED MANUSCRIPT

T
IP
CR
Fig. 3 - The deformation map

In our case, the frame acts as the skeleton and the DOFs of the nodes act as the transformations moving

US
and deforming the beams (i.e. the bones). While, in the case of skinning, simple rotation matrices are
enough to describe how the bones move, in our case there is the need to find a more complex
mathematical representation of the bone motions. Indeed, these latter do not have only rigid body motion:
they can translate, rotate, bend, shrink, twist, etc. It is known that the deformation of any deformable
AN
object can be described by a deformation map, that is, a function which maps any point of the object in its
initial state, to the same material point in the deformed state, as illustrated in Fig. 3. This holds particularly
for beams, when one needs to find its 3D deformed shape. With the deformation map associated to each
M

beam, a step towards controlling the voxelized object’s deformation with the frame’s deformation can be
achieved.
ED

M
Z

Y
PT

M(X,Y,Z)
CE

1
X
2
AC

M’(x,y,z)

Fig. 4 - Beam's deformation map

We consider a beam oriented into the positive x direction as shown in Fig. 4, with its centerline extending
from node 1 to node 2. M(X,Y,Z) is an arbitrary point of the beam in the initial state. In the deformed

7
 ACCEPTED MANUSCRIPT

state, this point is located at M’(x,y,z). All the aforementioned coordinates are expressed in the beam local
coordinate system as depicted in Fig. 4. It can be shown [26] that:
𝑢𝑥 (𝑋) 0
𝑀′ (𝑥, 𝑦, 𝑧) = 𝜙(𝑀) = 𝑋𝑒⃗⃗⃗⃗𝑥 + [𝑢𝑦 (𝑋)] + Λ(𝑋) [𝑌] (8)
𝑢𝑧 (𝑋) 𝑍

Where 𝑢𝑖 (𝑋) is the translational DOF along the 𝑖–axis at the position 𝑋. Λ(𝑋) = 𝑓 (𝜃𝑥 (𝑋), 𝜃𝑦 (𝑋), 𝜃𝑧 (𝑋))
(with 𝜃𝑖 (𝑋) being the rotational DOF around the 𝑖–axis at the position 𝑋) is a tensor that characterizes the

T
rotation in space of a cross section to which M belongs. More explicitly Λ(𝑋) is expressed as:

IP
Λ(𝑋) = 𝑅𝑧 (𝜃𝑧 (𝑋)) ∙ 𝑅𝑦 (𝜃𝑦 (𝑋)) ∙ 𝑅𝑥 (𝜃𝑥 (𝑋)) (9)

CR
Where 𝑅𝑖 being the matrix of a rotation about the 𝑖–axis.
With the frame’s deformation computation the values of the DOFs are only known at the nodes 1 and 2, so

US
we used shape functions to extrapolate their values at any location along the beam (to reduce clutter the
dependency on 𝑋 has been omitted):
AN
𝑢𝑥 𝑁1 0 0 0 0 0 𝑁2 0 0 0 0 0
𝑢𝑦 0 𝑁3 0 0 0 −𝑁4 0 𝑁5 0 0 0 −𝑁6
[𝑢 ] = [ ]𝑈
𝑧 0 0 𝑁3 0 𝑁4 0 0 0 𝑁5 0 𝑁6 0
𝜃𝑥 0 0 0 𝑁1 0 0 0 0 0 𝑁2 0 0
M

(10)
𝑑𝑢𝑧
𝜃𝑦 = −
𝑑𝑋
𝑑𝑢𝑦
ED

𝜃𝑧 =
𝑑𝑋
PT

Where: 𝑈 = [𝑢𝑥1 𝑢𝑦1 𝑢𝑧1 𝜃𝑥1 𝜃𝑦1 𝜃𝑧1 𝑢𝑥2 𝑢𝑦2 𝑢𝑧2 𝜃𝑥2 𝜃𝑦2 𝜃𝑧2 ]𝑇 and:

1
𝑁1 (𝑋) = − (𝑋 − 𝑋2 )
CE

𝐿
1
𝑁2 (𝑋) = (𝑋 − 𝑋1 )
𝐿
3𝑥̅ 2 2𝑥̅ 3
AC

𝑁3 (𝑋) = 1 − 2 + 3
𝐿 𝐿
2𝑥̅ 𝑥̅ 2 , 𝑥̅ = 𝑋 − 𝑋1 (11)
𝑁4 (𝑋) = 𝑥̅ (−1 + − )
𝐿 𝐿
2
𝑥̅ 2𝑥̅
𝑁5 (𝑋) = 2 (3 − )
𝐿 𝐿
𝑥̅ 2 𝑥̅
𝑁 6 (𝑋) = (1 − )
{ 𝐿 𝐿

Equations (8-11) are all written in local coordinate system (LCS). On implementing them, they have been
adapted to handle points coordinates in GCS and yield points in deformed state M’ coordinates in GCS as
well; an independence to the beam orientation has also been implemented. On a computational aspect,

8
 ACCEPTED MANUSCRIPT

morphing a parallelepiped beam according to the deformation of the underlying centerline was made by
first meshing the beam surface and then moving the mesh’s vertices by using the associated deformation
map.
With each beam of the frame associated to a deformation map, there is one more step towards deforming
the whole object. This is where skinning [22] has been harnessed to the proposed modeling framework. In
our case, morphing the voxelized object according to the underlying skeleton frame is made through five
steps:
1. Voxel mesh’s densification – By default a voxel only has 8 vertices (its corners). Moving only
these 8 vertices is not enough to accurately capture its actual deformed shape. That is why the

T
number of vertices defining its shape must be increased by re-meshing its surface. The mesh

IP
density is therefore increased to 5-10 faces along each direction, which is high enough for an
accurate mesh deformation.

CR
2. Extraction of the vertices set – All the voxels’ vertices are extracted and stored in a list without
duplicates.
3. Weights computations – The vertices should move according to which beams they are (likely to

US
be) influenced by. Therefore, we first define an influence zone for each beam, a zone which is the
space occupied by the two voxels that the beam is connecting, any vertex belonging to this space
is then influenced by the beam deformation. A vertex at the border between many influence zones
AN
is considered to belong to all these influence zones. Using these assumptions, all the beams
influencing any vertex can be found. We consider each vertex to be equally influenced by its
1
influencers, which yield the weights: 𝜔𝑖 = 𝑁 for all the beams, where N is the number of beams
M

influencing that vertex.

4. Vertices motions – Each vertex is then moved by using a weighted average deformation map
ED

(which is a blend of the influencing beams’ deformation maps).

5. Voxels shape update – Finally each voxel shape is updated by using the new position of its
vertices.
PT

On a computational aspect, as moving any vertex is independent of the others locations, the process for
deforming the voxelized object is friendly to parallelization, and hence speed.
CE

2.4 Modeling scheme validation

2.4.1. Homogeneous structures simulations compared with Finite Element Analysis

AC

While the proposed model of continuum mechanics modeling scheme is used as a tool for conceptual
design (to rapidly know how a distribution qualitatively behaves), we found it worth gaging its accuracy
compared to finite element method (FEM) simulations. Here two cases have been used to evaluate the
modeling scheme:

- A cantilevered thick beam (50mm10mm10mm) made of a single material (E = 20GPa, G =7.69

GPa - structural steel) loaded at its free end with a 800N force.
- A thin squared plate (30mm30mm2mm) also made of a homogeneous MD (E=1000MPa,
G=385MPa), fixed on the sides and loaded on a little square (3mm3mm) at its center with a
pressure of 16.5MPa.

9
 ACCEPTED MANUSCRIPT

For these two cases, the maximum displacement was used as a measure for comparison. For the beam a
voxel size of 2mm was chosen, leading to a voxelized object of 2555 voxels. The same voxel size was
used for the plate which is then made of 15152 voxels. The deformed shapes of the two cases are shown
in Fig. 5. The FEA was run using the commercial software Comsol Multiphysics®; free tetrahedral
elements of the predefined size “fine” were used. The results of the simulation are outlined in Table 1.
Table 1 - Maximum displacement (mm) for the two cases

Proposed voxel- Finite Element

Case based model Analysis
Beam 1.86 2.03

T
Thin plate 3.7 3.6

IP
CR
(a) (b)

US
AN
M
ED

Fig. 5 - Two simulation cases with our modeling scheme: (a) cantilevered beam loaded at its free end, and (b)
PT

thin plate loaded fixed on the sides and at its center

For these two cases, the deformations are physically realistic and maximum displacements are within the
ranges of the results obtained from finite element analysis (FEA) with an extremely fine mesh. This
CE

discrepancy from the results yielded by a FEA may have multiple causes. The two most plausible are:
loads, boundary conditions (BCs) locations and voxel size (discretization) which are not independent. For
the cantilevered beam case simulated in a FEA software, the force was applied as a force per unit area on
AC

the beam free end surface, and a fixed constraint (all DOFs set to zero) was set on the fixed end surface. In
the proposed voxel-based modeling scheme, the force was equally distributed over the (25) voxels centers
located at the free end, while the voxels centers located at the other end were fixed. The distance between
where the beam is actually fixed and where the load is actually applied is shorter than in the real case
simulated in the FEA software, therefore the beam actually simulated with our modeling scheme is stiffer
than the actual one hence a lower maximum displacement. Furthermore, our modeling scheme does only
support point loads (and moments) applied at the underlying frame’s nodes; this is not the type of load
which has been used in the FEA. Finally, as the voxel size decreases, BCs and loads get closer to their
actual locations and their distributions get more accurate. In the beam case, we used a voxel size of 2mm
which means BCs and loads locations are 1mm off their actual locations inwards the beam. Again, the
proposed modeling scheme is not meant to be as accurate as established methods like FEM but is more
aimed to provide qualitative evaluation of materials distributions in conceptual design.

10
 ACCEPTED MANUSCRIPT

2.4.2. Qualitative evaluation for heterogeneous structures simulation

The capability of our modeling scheme to handle multi-material simulation has also be somewhat gaged.
We used a cantilevered beam with increasing stiffness along the width, the MD is such that five materials
with Young moduli (all expressed in GPa) 0.08, 0.1, 0.8, 10 and 20 were arranged along the beam width
as shown in Fig. 6.a. The beam is loaded at its free end with a 800N force. As one could expect stiffer
sections would deform less since they are more load resistant; our model is able to reproduce this behavior
as shown in Fig. 6.b.

T
(a) Young’s modulus (b)

IP
20GPa
10GPa

CR
0.8GPa

0.1GPa
0.08GPa

US
Perspective view
AN
M
ED

Perspective view

Right view
PT

Fig. 6 – Multi-material beam with varying materials width wise

Another multi-material beam case has been simulated. We used a cantilevered beam with in the middle a
CE

material ways softer than the material of the remaining of the beam. A load oriented upwards was applied
to its free end. As one could expect, the beam bends at the soft section (see Fig. 7).
AC

Materials distribution Simulation

Soft material

Stiffer material

Fig. 7 - Cantilevered beam with a soft and hard materials distribution

11
 ACCEPTED MANUSCRIPT

3. Modeling the stimulus responsive behavior of smart materials

With the discovery of the shape memory effect and piezoelectricity, the range of SMs properties has
significantly expanded, both in terms of stimuli and responses to these stimuli. While the main purpose of
this section is not to provide a review of SMs, its endeavor to empower designers – which are likely to be
non-experts of SMs – to use them, make it valuable to outline the SMs realm. Such a review would also
direct how the stimulus responsive behaviors of these materials can realistically be modeled in the scheme
we are using. Many SMs definitions exist in the literature, however all of them are in agreement with the
fact that these are materials whose states are altered as a response to a specific change – a stimulus – in
their environment. Behind such properties, can be seen the research endeavor to incorporate intelligence in

T
the matter so that it behaves autonomously by sensing, reacting and adapting to the environment, as does

IP
any biological system.
Depending on how they respond to a stimulus, SMs may fall in any of the following groups:

CR
- Shape changers: these are those which respond to stimuli by strain or stress. While some of them
simply exhibit change in size (e.g. hydrogel, piezoelectric material, etc.) others, such as shape
memory materials, react by changing shape.

US
- Optical sensors: within this group are materials whose response is optically perceivable; this
includes for instance thermochromic materials, triboluminescent materials or switchable mirrors.
- Converters: these materials are those whose response is typically a signal that can be used as a
AN
stimulus for another SM or to provide information about a medium’s state. Examples of such
materials include piezoelectric material, thermoelectric material or photovoltaic material.
- State changers: SMs usually have a single condition; state changers are those whose conditions
M

change in response to the right stimulus. Examples of these are electro-/magneto-rheological

fluids or shear thickening fluids.
ED

This categorization is the result of an analysis of the world of SMs as presented by Lefebvre, et al [27].
We consider as SMs, materials which sense and react to stimuli at their own, that is, which do not need
another material to perform such functions. This excludes for instance dielectric elastomers [28], because
PT

even though they do shrink and expand in response to a potential difference, they are formed by the
combination of the dielectric material sandwiched between two compliant electrodes. And the force
causing this deformation is the electrostatic force attracting the electrodes together. In other words
CE

dielectric elastomers (sometimes referred to as electrostrictive materials) are not sensitive to an electric
field without electrodes.
AC

Our proposed approach has the potential to be used for a broad range of SMs; but for the purpose of
illustrating the fundamental concept, we chose shape changers. The use of these latter to achieve a desired
effect is quite challenging. Indeed as explained in the introduction, depending on how they are spatially
arranged within a part, they can lead to totally different behaviors. Finding a right distribution to achieve a
desired shape change is more challenging, for instance, than finding a distribution to achieve a visual
effect. For some of the SMs which – in the current state of the art – work in liquid form and/or in
homogeneous states, designing a MD might not be of interest. This is the case of converters and state
changers. Therefore, as mentioned in the introduction, this paper is concerned with the shape changers
group.

12
 ACCEPTED MANUSCRIPT

3.1 Description of shape changing materials and materials’ properties to be considered

Within the group of shape changing materials (SCMs), it is worth distinguishing two subgroups [29]:
programmable SCMs (p-SCMs) and non-programmable SCMs (np-SCM). In np-SCMs the way the
material can change shape is predefined during fabrication and the shape change is limited to dimensions
change (shrinking or expansion isotropically or not). p-SCMs change shape according to a pathway that
can be altered regardless of how they were manufactured. In other words, after manufacturing they can be
trained to shift shape between almost any shapes. These materials are typically shape memory materials
[29, 30]. Basically, they are formed into one permanent shape, and can be thermo-mechanically trained (or
programmed) to assume one (or several) temporary shape(s) from which the permanent shape can be

T
recovered. As such achieving a specific shape change with p-SCMs is rather a thermomechanical training
issue than a material spatial arrangement determination issue. That is why this work is focused on smart

IP
materials distribution with np-SCMs, which are described in the following paragraphs. The descriptions
serve the purpose of briefly introducing the reader to them and most importantly they aim at deriving

CR
equations governing the materials’ responses for our modeling scheme. First of all, we derive the stiffness
equations for a generic member (or beam) made of a np-SCM.
For conventional (inert) materials, the beams’ DOFs are governed by the member stiffness equations:

US
̅ 𝑢̅ = 𝑓𝑀̅
𝐾 (12)
̅
Where 𝑓𝑀 is the vector containing all the (external) mechanical forces and moments applied to the beam
AN
(at its nodes). Equation (12) basically means that the only way for the beams to deform is by the
mechanical forces and moment applied at their nodes. In our modelling scheme, the action of the smart
materials will be modeled by introducing initial force effects, in a way similar to how thermal forces are
M

modeled:

̅ 𝑢̅ = 𝑓𝑀̅ + 𝑓 ̅(𝑺)
𝐾 (13)
ED

𝑺 denotes the stimulus. For simplicity in understanding we will consider a planar bar member. In an
unloaded state and upon exposure to the stimulus, its length is free to change from 𝐿 to 𝐿 + 𝐿, where 𝐿
is a function of the stimulus and the material properties:
PT

𝐿 = 𝐿 × 𝑔(𝛂, 𝑺) (14)
CE

𝛂 denotes a vector containing all the SMs properties related to its stimulus responsive behavior. The strain
due to the stimulus can then be expressed as:
𝐿
AC

𝜖𝑆 = = 𝑔(𝛂, 𝑺) (15)
𝐿
𝐹
Now let’s assume that the bar is also subjected to an (axial) force F causing the stress 𝜎 = 𝐴 (with A being
𝜎
the bar’s cross-section), which induces the strain 𝜖𝑀 = 𝐸 . The total strain in the bar is then:

𝑢̅𝑥𝑗 − 𝑢̅𝑥𝑖 𝜎
𝜖= = 𝜖𝑀 + 𝜖𝑆 = + 𝑔(𝛂, 𝑺) (16)
𝐿 𝐸
Which can be rewritten as follows:
𝐸𝐴
(𝑢̅𝑥𝑗 − 𝑢̅𝑥𝑖 ) = 𝜎𝐴
⏟ + 𝐸𝐴𝑔(𝛂, 𝑺)
⏟ =𝐹
𝐿 𝑚𝑒𝑐ℎ𝑎𝑛𝑖𝑐𝑎𝑙 𝑖𝑛𝑡𝑒𝑟𝑛𝑎𝑙 𝑓𝑜𝑟𝑐𝑒
(17)
𝑎𝑥𝑖𝑎𝑙 𝑓𝑜𝑟𝑐𝑒 𝑖𝑛𝑑𝑢𝑐𝑒𝑑 𝑏𝑦 𝑡ℎ𝑒 𝑠𝑡𝑖𝑚𝑢𝑙𝑢𝑠

13
 ACCEPTED MANUSCRIPT

Where 𝐹 denotes the total internal force. The joint forces are related to F as:

[𝑓𝑥̅ 𝑖 𝑓𝑦̅ 𝑖 𝑓𝑥̅ 𝑗 𝑓𝑦̅ 𝑗 ] = ⏟ ̅

[𝑓𝑀𝑥𝑖
̅
𝑓𝑀𝑦𝑖
̅
𝑓𝑀𝑥 𝑗
̅ ]+⏟
𝑓𝑀𝑦𝑗 𝐸𝐴𝑔(𝛂, 𝑺) ∙ [−1 0 1 0]
(18)
̅
𝑓𝑀 𝑓𝑆̅
The member stiffness equations as:
𝑢̅𝑥𝑖
1 0 −1 0
𝐸𝐴 0 0 0 0 ̅ 𝑦𝑖 𝑢
[ ] = 𝑓𝑀̅ + 𝑓𝑆̅ (19)
𝐿 −1 0 1 0 𝑢̅𝑥𝑗
0 0 0 0 [𝑢̅𝑦𝑗 ]

T
Regarding the beam elements used in our modeling scheme, the internal force due to the stimulus will be
expressed as:

IP
𝑓𝑆̅ = 𝐸𝐴𝑔(𝛂, 𝑺)[−1 0 0 0 0 0 1 0 0 0 0 0]𝑇 (20)

CR
What will differentiate the various modeled SMs will be the expression of 𝑔(𝛂, 𝑺) and the conditions of
application of the related force.

US
Piezoelectric material
Of interest in the scope of this paper is the reverse piezoelectric effect by which a voltage (or equivalently
AN
an electric field) generates strain. An inherent electrical property of these materials is what is called the
polling direction, usually referred to as the 3-direction; it is the direction along which most of the electric
dipoles within the material are oriented. The other two orthogonal directions are denoted 1 and 2. At
M

constant stress, strain generated by an electric field is defined by the equation [31]:

𝜖𝑗 = 𝑑𝑖𝑗 𝐸𝑖 (21)
ED

Where:
- 𝐸𝑖 , is the electric field in direction 𝑖 ∈ ⟦1,3⟧;
PT

- 𝜖𝑗 , is the generated strain in direction 𝑗 ∈ ⟦1,6⟧, with the convention 4: 1-2, 5: 1-3, 6: 2-3 for the
shear strains;
- 𝑑𝑖𝑗 , are piezoelectric strain coefficients.
CE

Commonly used piezoelectric coefficients for actuation are 𝑑33 (strain along the polling direction induced
by an electric field along that same direction) and 𝑑31 (strain along directions perpendicular to the polling
AC

direction by electric field’s component along that same direction). Function 𝑔(𝛂, 𝑺) as introduced in
equation 14 is then expressed as:

𝑔(𝛂, 𝑺) = 𝑑3𝑗 𝐸3 (22)

In our modelling scheme piezoelectric material properties will be represented by the coefficients 𝑑33 and
𝑑31 ,they will also be characterized by a polling direction (x, y or z). The internal force due to the stimulus
will depend on the beams orientation: for beams along 𝐸3 , 𝑑33 will be used for beams in perpendicular
directions 𝑑31 will be used. This setup assumes then that only situations where the electric field is along
the material polling direction can be simulated. The highest strains are in fact obtained in this situation.
Electrostrictive material

14
 ACCEPTED MANUSCRIPT

Dielectric materials are materials that do not conduct electricity, nevertheless they are responsive to
electric field by (among others) exhibiting electrostriction. These materials are made of electric domains
which are randomly oriented within the material. When a sample is subjected to an electric field, the
electric domains get polarized along the electric field. As the opposite sides of these domains are then
charged with opposite charges they attract each other, thus they shrink in the field direction and the
elongate in perpendicular directions according to the material Poisson’s ratio. The effect is a second order
one, that is, the resulting deformation is proportional to the square of the electric field; particularly
reversing the field does not change the sign of the strain. The so described mechanism is to be
distinguished from what is usually referred to as electrostriction in dielectric polymers [28] and which is
discarded from the scope of this work as explained in section 3 introduction.

T
In a stress free sample of the material, three electrostriction coefficients [32] can be defined to relate the

IP
induced deformations to the polarization:

CR
𝜖1 𝑄11 𝑄12 𝑄12 0 0 0 𝑃12
𝜖2 𝑄12 𝑄11 𝑄12 0 0 0 𝑃12
𝜖3 𝑄12 𝑄12 𝑄11 0 0 0 𝑃12
𝜖4 = 0 0 0 𝑄44 0 0 𝑃2 𝑃3
(23)

US
𝜖5 0 0 0 0 𝑄44 0 𝑃3 𝑃1
[𝜖6 ] [ 0 0 0 0 0 𝑄44 ] [𝑃1 𝑃2 ]
Where 𝑃 = [𝑃1 𝑃2 𝑃3 ]is the polarization vector. In case of induced polarization (which is the case when
AN
the material is subjected to an electric field), there is a single polarization direction, the one in the
direction of the applied field. This direction is denoted 1, which zeros P2 and P3 from equation 23. The
polarization is related to the electric field by: 𝑃1 = 𝜖𝐸1 , where 𝜖 is the static dielectric constant of the
M

material. Equation 23 can then be rewritten as:

𝜖1 = 𝑄11 𝜖 2 𝐸12 = 𝑀11 𝐸12

ED

{𝜖2 = 𝑄12 𝜖 2 𝐸12 = 𝑀12 𝐸12 (24)

𝜖3 = 𝑄12 𝜖 2 𝐸12 = 𝑀12 𝐸12
PT

Function 𝑔(𝛂, 𝑺) will then be written as:

𝑔(𝛂, 𝑺) = 𝑀1𝑗 𝐸12 (25)

CE

Electrostrictive materials properties will then be represented by the coefficients 𝑀11 and 𝑀12 . The
material polarization direction will be defined as the applied field’s direction 𝐸1 . For beams along 𝐸1 , 𝑀11
will be used and for those in transverse directions 𝑀12 will be used.
AC

Magnetostrictive materials
Any ferromagnetic (e.g. nickel, iron etc.), ferrimagnetic (e.g. iron II,III oxide) and antiferromagnetic (e.g.
chromium, nickel oxide, etc.) material exhibits a phenomenon called magnetostriction. Once subjected to
a magnetic field they exhibit a slight change in dimension. Roughly speaking this is due to rotation of
magnetic dipoles within the material as they align with the applied magnetic field. There is a positive
magnetostriction where the material elongates along the applied field, and a negative magnetostriction in
which the material shrinks. For instance when exposed to a strong magnetic field iron can elongate by
0.002 %, while nickel contracts by 0.007%. Some of these materials are termed as “giant magnetostrictive
materials” in reference to the higher strains they can produce. This is the case of Terfenol-D (up to 0.1%)
or NiMnGa alloys (up to 9%). There are also dimensions’ changes in the directions perpendicular to the

15
 ACCEPTED MANUSCRIPT

∆𝐿
applied field. The effect is measured by the developed strain: 𝜆 = = 𝑓(𝐻) . H being the applied
𝐿
magnetic field. Similarly to electrostriction, it is a second order effect: reversing the applied magnetic field
doesn’t change the developed strain. Magnetostrictive materials are usually characterized [7] by their
strain at magnetization saturation λs (which is the maximum strain – or minimum strain in case of negative
magnetostriction – that can be developed by the material) and the saturation magnetic field Hs. In the
region before saturation, the dependency of λ to H is quadratic.
In our modeling scheme, we will only consider the strain developed along the applied field (as it is the
most significant) and values of magnetic field will be considered to be within the linear region before
saturation. The material’s magnetostrictive properties will be λs and Hs. Relation between magnetostriction

T
and magnetic field can then be approximated by a linear law:

IP
𝜆𝑠
|𝐻|, |𝐻| < 𝐻𝑠
𝑔(𝛂, 𝑺) = 𝜆 = {𝐻𝑠 (26)

CR
𝜆𝑠 , |𝐻| ≥ 𝐻𝑠

US
Only beams along the applied field will be subjected to the internal force resulting from magnetostriction.

Photostrictive materials
AN
Photostrictive materials are materials whose dimensions change when exposed to light, a change which is
different from and more important than the one associated with heat induced by the light. Photostriction is
M

found in four main types of materials [33] including ferroelectric materials, polar and non-polar
semiconductors, and organic polymers. Some photostrictive materials shrink while others expand. The
mechanism responsible of the phenomenon is quite different depending on the material: in ferroelectric
ED

material photostriction is due to a combination of photovoltaic and the reverse piezoelectric effect,
whereas in organic polymers the phenomenon is due to photoisomerization (light induced change in
∆𝐿
molecule structure). The effect is usually quantified with a single measure of strain, usually denoted
PT

𝐿
and referred to as photostriction coefficient, which is an indication that the behavior is isotropic. While in
ferroelectric, photostriction coefficient of 0.45% is deemed as a giant photostrictive response, in nematic
CE

elastomers photostriction can be up to 400% [34]. The literature on these materials is populated with
experimental data showing the developed strain versus light’s wavelength [33], exposure time to light [34]
and light’s intensity [35] but there are no close form relationships between the electrostriction coefficient
and these characteristics of the stimulus. In addition there is a dependence on light penetration depth. In
AC

our modeling scheme we elected to restrain the dependency of photostriction to light’s intensity (as the
easiest controllable parameter), the effect is modeled as:
∆𝐿
= 𝑘𝐼 (27)
𝐿
Where, I is light’s intensity and k, a material property. This relationship is based on the experimental work
reported in [35] (especially Figure 2.c of this paper). In addition a parameter for light penetration depth is
introduced, this will tell the number of voxels (thus the number of beams) in the material thickness that are
reached by light.

16
 ACCEPTED MANUSCRIPT

Hydrogel
Hydrogels – which only work in wet conditions – are able to absorb or repel water (in a way similar to a
sponge, except that they can retain the absorbed water in such a way that even pressure on it, may not
release the water), and thus they can drastically change volume. Depending on their chemical composition
they can be responsive to heat (as in most encountered hydrogels), light, electricity, solution properties
(pH, salinity, concentration of a specific constituent, etc.). The exhibited shape change is an isotropic
either shrinkage or expansion, depending on the chemical composition. In the case of heat as stimulus,
there is lower threshold by which the dimension’s change begins (usually referred to as lower critical
solution temperature – LCST). Away from the critical temperature, the change stops, as the material

T
reaches an equilibrium state. The material behavior is usually tracked by a volumetric swelling ratio
(VSR), whose definition varies according to authors. In [21], the following definition of VSR was used:

IP
𝑉0
𝜈𝑠 = (28)

CR
𝑉𝑠
Where 𝑉0 is the material’s volume in dry state, and 𝑉𝑠 the current material’s volume in swollen state. As
such: 0 < 𝜈𝑠 ≤ 1. With this definition, the higher is 𝜈𝑠 , the lower is the material’s volume. In the case of
shrinkage of the material with rising temperature, VSR’s dependency on temperature is expressed [21] as:

US
𝐿𝐶𝑆𝑇 − 𝑇 −1
𝜈𝑠𝑚𝑖𝑛
𝜈𝑠 (𝑇) = +(𝜈𝑠𝑚𝑎𝑥 − 𝜈𝑠𝑚𝑖𝑛 ) [1 +exp ( )] (29)
𝑘
AN
Where T is the temperature, k is a constant controlling how gradual the transitional behavior around the
LCST is, 𝜈𝑠𝑚𝑖𝑛 and 𝜈𝑠𝑚𝑎𝑥 are the limits of the swelling ratio in fully swollen and collapsed (dry) state
respectively. As the dimension change is isotropic (strain resulting from the volume change is the same in
M

all the directions), function 𝑔(𝛂, 𝑺) can be expressed as:

1
𝑔(𝛂, 𝑺) = [𝜈𝑠 (𝑇)]−3 − 1 (30)
ED

Any beam in our modeling scheme made of hydrogel will then be subjected to the internal force resulting
from temperature-driven volume change.
PT

3.2 Stimulus modeling

CE

The stimuli triggering the aforementioned SMs are: heat, light, electric field or equivalently voltage, and
magnetic field. These can clearly be separated in scalar stimuli and vector stimuli. In our modeling
scheme a stimulus is considered as an environment variable that is sensed by each voxel (and thus each
AC

beam) the same way, that is, the value of the stimulus is considered to be the same for all the voxels.
While such assumption is valid for electric or magnetic field (considering no electromagnetic shield is to
be modeled), for heat and light it is questionable. Indeed heat propagates (from hot regions to colder ones)
and light gets absorbed as it propagates through a medium. As the proposed modeling scheme is meant to
be used in conceptual design (where the focus is more on functionality than accuracy), we elected not to
take these phenomena into account. Nevertheless the framework may be extended to model all the stimuli
field in the simulated object more accurately.

17
 ACCEPTED MANUSCRIPT

4. Validation with predefined distributions

In order to ascertain the accuracy of the proposed modeling framework for SMs modeling, a few existing
MDs have been simulated. Namely the printed smart valve from [36] and the hypothetical actuators from
[21]. All these examples are based on heat responsive hydrogel.

4.1 Smart hydrogel valve

Bakarich et al. [36] have demonstrated 4D printing with a smart valve. Basically the printed valve is able
to regulate water flow according to this latter’s temperature. The SM responsible of this behavior is a

T
thermally responsive poly (N-isopropylacrylamide) hydrogel commonly referred to as PNIPAAm. Its
LCST (cf. subsection 3.2.1) is between 32°C and 35°C. The material exhibits a large decrease in water

IP
content as the temperature is around LCST. The valve, as depicted in Fig. 8, has been printed with 2
materials: an epoxy based adhesive (Emax 904 Gel-SC) for the inert sections and hydrogel for the active

CR
sections. As water flows through the central tubing (from the top) and warms the actuating hydrogel strips,
these shrink to close the outlet, thus blocking the water flow.
Materials’ properties were used as measured in [36, 37] and are summarized in Table 2.

US
Table 2 - Material properties used for the smart valve

E (MPa) G (MPa) 𝜈𝑠𝑚𝑖𝑛 𝜈𝑠𝑚𝑎𝑥 LCST (°C)

AN
Emax 2.7 0.96
Hydrogel 1 0.35 0.98 0.4 33.5
M

The valve has been modeled with a voxel’s size of 1mm. Voxels in the first row from the top were all
fixed. Results of the simulation for an increasing temperature are shown in Fig. 8.
ED

(a)
PT
CE

20 C 25 C 30 C 31 C 31.5 C
(b)
AC

(c)

Top Top

Fig. 8 – (a) Results of the simulation of the smart valve at various temperature – (b) Side view at 20°C – (c)
Side view at 31.5°C

The shape change pattern yielded by our model is quite similar to the one exhibited by the actual printed
valve as shown in [36]. The hydrogel strips clearly shrinks isotropically and pull the bottom inert section
towards the tubing outlet.

18
 ACCEPTED MANUSCRIPT

4.2 Hydrogel actuator

In the previous example, the hydrogel sections work mostly in homogeneous configurations (the strips)
and as such the motion they are responsible of are limited to one dimensional motions. This is a typical
case in hydrogel applications. Using FEM based simulations Westbrook and Qi [21] designed hydrogel
actuators that leverage heterogeneities to more complex shape changes. One of their theoretical actuators
has (partially) been modeled and simulated with our voxel-based framework. It is the in which hydrogel
shrinking is converted into bending. Such motion has been realized by patterning the hydrogel (red) into
another non responsive hydrogel (blue) as shown in Fig. 9. Materials’ properties were used from their
study (some of them were estimated). The simulated actuator is simulated with the left end (see Fig. 9)

T
fixed.

IP
Initial shape

CR
US
AN
25 C 30 C 30.5 C 31 C 32 C
M

Fig. 9 - Hydrogel actuator

The hydrogel (red) sections do shrink in the three dimensions as expected and the generated eigenstrain
clearly induces a bending motion. Nevertheless there is a discrepancy between the way the actuator bends
ED

in our model and theirs. In both models the actuator bends upwards. In our modeling scheme, as the
temperature increases sections near the fixed end bends downwards (without a curvature change). This
bending motion is progressively reversed as one moves away from that end. This downward bending
PT

could be explained as an accommodation to the fixed end. In order to ascertain the actual behavior the
actuator should be printed.
CE

5. Conclusions and future work

AC

In this paper a contribution has been made to SMs modeling and simulation for conceptual design. A
voxel-based modeling framework has been put forth to model and simulate the behavior of both
conventional materials and (non-programmable) shape changing SMs (np-SCM). With this modeling
framework, the possibility is given to designers to rapidly test a given distribution of SMs and check how
it behaves upon exposure to stimulus before proceeding into detail design. Chunking matter into voxels
was motivated by the need to pattern different materials within a geometry and in any spatial arrangement.
Another rationale behind this approach is to give designers the ability to benefit the material complexity
capability now affordable by AM.
In the modeling scheme, the main computation occurs not on the voxels themselves but on the underlying
frame formed by beams connecting the voxels’ centers. The resulting frame’s nodes’ DOFs are then used
to compute a deformation map for each beam and, using skinning (a set of computer graphics techniques

19
 ACCEPTED MANUSCRIPT

for animating characters), the voxels are morphed. Simple models of SMs, in the form of internal forces
changing the beams’ lengths, have been introduced. These models are based on the actual behaviors of
SMs including piezoelectric material, electro-/magneto-/photostrictive materials and hydrogel. The
modeled materials are quite representative of this class of SMs. Furthermore, the methodology for
deriving these materials’ models can readily be used to derive models for others SMs.
A few simulations were run with the proposed modeling scheme in order to somehow validate it, both as
regards conventional materials and SMs behaviors. The test cases used to gage the simulation engine
include homogenous beam and plate, and heterogeneous beams (with materials of varying stiffnesses), as
regards the conventional material behavior. For SMs, a printed smart hydrogel valve [36] and theoretical

T
hydrogel actuators from [21] were simulated. All these simulation were quite fast and in pretty good
agreement with either experiment or FEA. Worth highlighting nevertheless, is that the proposed voxel-

IP
based modeling and simulation scheme is not meant to be a surrogate to established methods such as
FEM; rather it is a complimentary design tool mainly for conceptual design that precedes design efforts

CR
which can be invested in FEM. The goal is to rapidly specify any MD and check how it could behave.
In spite of the interesting results, our proposal still suffers from a number of limitations. The necessity to
anchor at least one voxel hinders the simulation of free objects. Besides, the way stimuli, especially the

US
scalar ones, are modeled typically ruled out the possibility to simulate deformations that could be induced
by non-uniform stimuli fields. Finally in the current setting, voxels making up an object must be of the
same size. An adaptive voxelization with a variable voxel size according to regions of interest could
AN
further reduce computation time and allow for bigger objects to be simulated.
In addition to the aforementioned limitations, future work can include effects such as collision, friction
M

and gravity. Furthermore as regards materials, shape memory materials could also be modeled; the
exploration of what a distribution including SMs of different types (e.g. one sensitive to heat and another
sensitive to light) is another way of extending the modeling scheme.
ED

From a practical perspective, a computational tool should embody the proposed modeling scheme (as
initiated in [38]) while allowing for conventional CAD modeling capabilities. It should also let designers
embrace the shape complexity capability peculiar to AM. Furthermore in addition to the given possibility
PT

of a seamless modeling and simulation of any MD (forward scheme), there should be – based on the
modeling scheme – a methodology for distribution computation, using a source and a target shape
(backward scheme). Such generated distribution could be a starting point for a finer design. These latter
CE

two future works are the subject of another publication by this group.
AC

CRediT authorship contribution statement

Germain Sossou: Conceptualization, Methodology, Writing – Original Draft. Hadrien Belkebir: Software.
Frédéric Demoly: Funding acquisition, Supervision, Conceptualization, Writing – Review & Editing. H.
Jerry Qi: Writing – Review & Editing. Samuel Gomes, Ghislain Montavon: Supervision.

Acknowledgements
This research activity is part of much larger project in the field of design for 4D printing. The authors
would like to thank the Ministère de l’Enseignement Supérieure et de la Recherche, the French

20
 ACCEPTED MANUSCRIPT

“Investissements d’Avenir” program, project ISITE-BFC (contract ANR-15-IDEX-0003) as main

financial supports of this research program, and S.mart Franche-Comté network for their participation.

References
1. Hull, C.W., Apparatus for production of three-dimensional objects by stereolithography. 1986,
Google Patents.
2. Vaezi, M., et al., Multiple material additive manufacturing – Part 1: a review. Virtual and
Physical Prototyping, 2013. 8(1): p. 19-50.

T
3. Stratasys. PolyJet Technology for 3D Printing. 03/21/2018]; Available from:
http://www.stratasys.com/polyjet-technology.

IP
4. Yamaguchi, K., et al., Generation of three-dimensional micro structure using metal jet. Precision
Engineering, 2000. 24(1): p. 2-8.

CR
5. Tibbits, S., 4D Printing: Multi-Material Shape Change. Architectural Design, 2014. 84(1): p. 116-
121.
6. Ferrara, M. and M. Bengisu, Materials that Change Color. PoliMI SpringerBriefs. 2014: Springer
International Publishing. 139.

US
7. Hristoforou, E. and A. Ktena, Magnetostriction and magnetostrictive materials for sensing
applications. Journal of Magnetism and Magnetic Materials, 2007. 316(2): p. 372-378.
8. Peterson, G.I., et al., 3D-printed mechanochromic materials. ACS Appl Mater Interfaces, 2015.
AN
7(1): p. 577-83.
9. Choi, J., et al., 4D Printing Technology: A Review. 3d Printing and Additive Manufacturing, 2015.
2(4): p. 159-167.
10. Leist, S.K. and J. Zhou, Current status of 4D printing technology and the potential of light-
M

reactive smart materials as 4D printable materials. Virtual and Physical Prototyping, 2016: p. 1-
14.
ED

11. Khoo, Z.X., et al., 3D printing of smart materials: A review on recent progresses in 4D printing.
Virtual and Physical Prototyping, 2015. 10(3): p. 103-122.
12. Momeni, F., et al., A review of 4D printing. Materials & Design, 2017. 122: p. 42-79.
13. Mulakkal, M.C., et al., Responsive cellulose-hydrogel composite ink for 4D printing. Materials &
PT

Design, 2018. 160: p. 108-118.

14. Baker, A.B., et al., 4D printing with robust thermoplastic polyurethane hydrogel-elastomer
trilayers. Materials & Design, 2019. 163: p. 107544.
CE

15. Thompson, M.K., et al., Design for Additive Manufacturing: Trends, opportunities,
considerations, and constraints. CIRP Annals - Manufacturing Technology, 2016. 65(2): p. 737-
760.
AC

16. Bodaghi, M., A.R. Damanpack, and W.H. Liao, Adaptive metamaterials by functionally graded
4D printing. Materials & Design, 2017. 135: p. 26-36.
17. Bodaghi, M., A.R. Damanpack, and W.H. Liao, Self-expanding/shrinking structures by 4D
printing. Smart Materials and Structures, 2016. 25(10): p. 105034.
18. Bodaghi, M., A.R. Damanpack, and W.H. Liao, Triple shape memory polymers by 4D printing.
Smart Materials and Structures, 2018. 27(6): p. 065010.
19. Ge, Q., et al., Multimaterial 4D Printing with Tailorable Shape Memory Polymers. Sci Rep, 2016.
6: p. 31110.
20. Noumenon. The material is the mechanism. 03/21/2018]; Available from:
http://noumenon.eu/#&panel1-12.
21. Westbrook, K.K. and H.J. Qi, Actuator Designs using Environmentally Responsive Hydrogels.
Journal of Intelligent Material Systems and Structures, 2008. 19(5): p. 597-607.

21
 ACCEPTED MANUSCRIPT

22. Jacobson, A., et al., Skinning: Real-time Shape Deformation. ACM SIGGRAPH 2014 Courses.
2014.
23. Biswas, A., V. Shapiro, and I. Tsukanov, Heterogeneous material modeling with distance fields.
Computer Aided Geometric Design, 2004. 21(3): p. 215-242.
24. Gupta, V. and P. Tandon, Heterogeneous object modeling with material convolution surfaces.
Computer-Aided Design, 2015. 62: p. 236-247.
25. Okereke, M. and S. Keates, Direct Stiffness Method, in Finite Element Applications: A Practical
Guide to the FEM Process. 2018, Springer International Publishing: Cham. p. 47-106.
26. Hjelmstad, K.D., The Linear Theory of Beams, in Fundamentals of Structural Mechanics. 2005,
Springer US: Boston, MA. p. 241-291.
27. Lefebvre, E., et al. Smart materials: development of new sensory experiences through stimuli

T
responsive materials. in 5th STS Italia Conference A Matter of Design: Making Society through

IP
Science and Technology. 2014. Milan, Italy: STS Italia.
28. Pelrine, R.E., R.D. Kornbluh, and J.P. Joseph, Electrostriction of polymer dielectrics with

CR
compliant electrodes as a means of actuation. Sensors and Actuators A: Physical, 1998. 64(1): p.
77-85.
29. Zhao, Q., H.J. Qi, and T. Xie, Recent progress in shape memory polymer: New behavior, enabling
materials, and mechanistic understanding. Progress in Polymer Science, 2015. 49–50: p. 79-120.

US
30. Mohd Jani, J., et al., A review of shape memory alloy research, applications and opportunities.
Materials & Design, 2014. 56: p. 1078-1113.
31. Ravichandran, G., Active Materials, in Springer Handbook of Experimental Solid Mechanics,
N.W. Sharpe, Editor. 2008, Springer US: Boston, MA. p. 159-168.
AN
32. Eyraud, L., et al., Matériaux électrostrictifs pour actuateurs. Revue de Physique Appliquee, 1988.
23(5): p. 879-889.
33. Kundys, B., Photostrictive materials. Applied Physics Reviews, 2015. 2(1): p. 011301.
M

34. Finkelmann, H., et al., A new opto-mechanical effect in solids. Phys Rev Lett, 2001. 87(1): p.
015501.
35. Zhou, Y., et al., Giant photostriction in organic-inorganic lead halide perovskites. Nat Commun,
ED

2016. 7: p. 11193.
36. Bakarich, S.E., et al., 4D Printing with Mechanically Robust, Thermally Actuating Hydrogels.
Macromolecular Rapid Communications, 2015. 36(12): p. 1211-1217.
PT

37. Bakarich, S.E., et al., Three-Dimensional Printing Fiber Reinforced Hydrogel Composites. ACS
Applied Materials & Interfaces, 2014. 6(18): p. 15998-16006.
38. Sossou, G., et al., Design for 4D printing: rapidly exploring the design space around smart
CE

materials. Procedia CIRP, 2018. 70: p. 120-125.

AC

22
 ACCEPTED MANUSCRIPT

CRediT authorship contribution statement

Germain Sossou: Conceptualization, Methodology, Writing – Original Draft. Hadrien Belkebir: Software.
Frédéric Demoly: Funding acquisition, Supervision, Conceptualization, Writing – Review & Editing. H.
Jerry Qi: Writing – Review & Editing. Samuel Gomes, Ghislain Montavon: Supervision.

T
IP
CR
US
AN
M
ED
PT
CE
AC

23
 ACCEPTED MANUSCRIPT

T
IP
CR
US
AN
M

Graphical abstract
ED
PT
CE
AC

24
 ACCEPTED MANUSCRIPT

Highlights

 Conceptual design framework for 4D printing is proposed

 Voxel based modeling and simulation of non-programmable shape changing smart materials
 Simulation of heterogeneous objects made of conventional materials and smart materials
 Fast evaluation of any material distribution behavior upon exposure to stimulus

T
IP
CR
US
AN
M
ED
PT
CE
AC

25