Professional Documents
Culture Documents
Additive Manufacturing
journal homepage: www.elsevier.com/locate/addma
Research paper
Keywords: This paper describes the research into 3D printed capacitive shear stress sensors. The context is the development
3D printed sensor of a technology that can potentially be used in shoe insoles. This is interesting in medical applications, such
Capacitive sensor as prevention of diabetic foot ulcers. The base geometry consists of a capacitive layout with parallel wire
Shear stress
structures and has two flexible elements on two sides. Experimental devices are optimized using electrostatic
Mechanical beam modeling
and mechanical models and printed using material extrusion additive manufacturing (AM) with a flexible
Electrostatic modeling
thermoplastic polymer (TPU). Two geometries with different sizes are designed and tested. The smallest design
is 8.75 × 8.75 × 11 mm and its responsivity is measured to be 0.9 pF∕MPa for a measurement range of ±140 kPa.
The design can easily be altered to meet specifications for a variety of other applications. The realized sensors
show some hysteresis, which we attribute primarily to the mechanical behavior. The capacitive measurement
of the displacement is quite linear. The mechanical hysteresis can partially be explained by the properties
of the printed material. The results are promising for application in shoe insoles and can relatively easily be
adapted to meet requirements for other applications.
∗ Corresponding author.
E-mail address: j.oprel@alumnus.utwente.nl (J. Oprel).
https://doi.org/10.1016/j.addma.2023.103674
Received 13 June 2022; Received in revised form 23 June 2023; Accepted 25 June 2023
Available online 4 July 2023
2214-8604/© 2023 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
J. Oprel et al. Additive Manufacturing 73 (2023) 103674
human foot the bandwidth is in a similar order of magnitude, a responsivity of 0.18 pF∕MPa and is able to measure shear forces in
typically 30–50 Hz [16–18]. two directions [30].
• Influences of other factors such as humidity changes, normal This research aims to provide a proof of principle, showing that
forces and other conductors in the near field should influence the shear stresses can be measured using a fully 3D printed capacitive
measurement to a minimum. sensor. In Section 2, electrostatic and mechanical models are discussed
• The surface area of the sensor should be small enough to prevent to study trends that can be expected for the behavior of such sensors.
underestimation of the peak shear stress [13], due to integration These models are used to guide the design of the sensing structures and
and averaging of stress over a larger area. For this reason a surface do not aim to describe the sensor behavior with high accuracy. The
area smaller than 10 × 10 mm2 [13] is desired. models help to choose an optimal design based upon requirements of
• The sensor design should pose no restrictions on the surface of
the sensor. In this research, two sensors are presented which have dif-
the sole. Bumps or other irregular shapes required by some sensor
ferent sensitivities and measurement ranges. The experimental methods
designs are not acceptable for the application, as these irregular
used for printing and characterization are discussed in Section 3 and the
shapes can form an additional risk factor for development of
obtained results are presented and discussed in Section 4.
DFUs.
Fig. 1. Sketch of the sensor design, with indication of the mechanical variables and parameters. The capacitive sensing structure is depicted in more detail in Fig. 2.
2
J. Oprel et al. Additive Manufacturing 73 (2023) 103674
Fig. 2. Sketch of the wire structure design of the sensor, with relevant parameters.
Fig. 5. Deviation of the hyperelastic beam model compared to a 3D FE model for the
Fig. 3. The rounded plate and wire structure of which the capacitances are compared. horizontal deformation at the end of a beam, for different beam thickness to height
ratios.
3
J. Oprel et al. Additive Manufacturing 73 (2023) 103674
33
2
³ responsivity
L d y in pF/m
32
´
1 dC
Electrostatic
per length
31
30
−1.6 −1.5 −1.4 −1.3 −1.2 −1.1 −1
y 0 = −1.246 Initial displacement y 0 in mm
1
x in mm
−1
−6 −4 −2 0 2 4 6
y in mm
Fig. 6. Top: responsivity-displacement behaviour of a wire structure (with parameters as in the large sensor design: 𝑆𝑦, in groups = 0.6 mm, 𝑆𝑦, groups = 4 mm, 𝑆𝑥 = 1.2 mm, 𝑑 = 0.4 mm).
The dashed line indicates which situation is plotted in the bottom graph. Bottom: top and bottom electrode structure with equipotential lines. (For an animated version of this
figure, the reader is referred to the web version of this article. The animation of this figure can also be seen digitally in some pdf-readers, such as Adobe Acrobat Reader DC.)
The beam model is based on the equilibrium equations derived of the remaining parameters are unknown, they are further optimized
by He et al. [35], in Appendix B. Suitable boundary conditions are concurrently. This is discussed in Section 2.4.3.
substituted for our specific design.
Alternatively, an FE method can be used to find the deformation of 2.4.1. Optimization of the initial displacement in the electrical design
beams or other structures. For this, the Neo-Hookean material model in Given a wire structure like sketched in Fig. 2, the initial dis-
®
COMSOL-Multiphysics is used. The materials are again modeled using placement (𝑦0 ) is optimized. Using the center-wire approximation, the
their initial shear modulus (𝜇) and incompressibility is assumed. displacement-capacitance behavior is studied. The main objective here
One of the advantages of the FE model is that the deformation is to maximize the mechanical responsivity, that is the capacitance
in the 𝑧-direction can be modeled, such that beams with a specific change per unit of relative movement of the bottom electrodes in the
width 𝐿 in the 𝑧 direction can be studied. This is in contrast to the 𝑦-direction. This responsivity is plotted for different values of the initial
beam model, where only 2-dimensional deformations are considered. displacement 𝑦0 in Fig. 6. It can be seen that there is an optimum for
The influence of this assumption can be studied by comparing the two the responsivity.
models. The results of this comparison are shown in Fig. 5. From the
Figure, it is clear that the beam model is only suitable for application 2.4.2. Optimization of the beam width
in certain ranges of beam widths 𝐿. For example, for thin beams with In the optimization of the beam width there is a trade off be-
a thickness to length ratio (𝐻∕𝑆b ) of 0.05, the deviation between the tween responsivity and influence of normal stress: a smaller beam
FE and the beam model is less than 5% if the beam width 𝐿 is at least width increases the displacement due to shear stress and, hence, the
half the beam length 𝑆𝑏 . Especially for beams with a very small responsivity is increased. However, this increased displacement results
( width)
or a combination of large width and large thickness to length 𝐻∕𝑆𝑏 in more influence of the normal force on both horizontal and vertical
ratio, the beam model does not provide an accurate description of the displacement of the structure.
deformation. In certain ranges (as depicted in Fig. 5), the hyperelastic The maximum normal and shear stress levels expected in this appli-
beam model is useful to find an estimate for the deformation, especially cation are 740 kPa and 140 kPa respectively [13]. That means that, with
since computation times are substantially lower than for FE models. the stress distributed over multiple beams, the normal and shear forces
on one beam (𝑃max and 𝑄max ) can be expressed as:
2.4. Optimization
𝑃max 𝑊tot
= ⋅ 740 kPa
𝐴c 𝑛beams 𝐻
Using the models established in the previous sections, the parame- (1)
𝑄max 𝑊tot
ters of the design as shown in Figs. 1 and 2 can be optimized to achieve = ⋅ 140 kPa
a large capacitance change per area of the sensor, whilst satisfying the 𝐴c 𝑛beams 𝐻
requirements as listed in Section 1. Whilst keeping all other parameters where 𝐴𝑐 is the cross-sectional area of a beam and 𝑛beams is the number
fixed, optimal values for the initial displacement (𝑦0 ) in the electrical of beams, which is 2 in the proposed design.
design and the beam thickness (𝐻) in the mechanical design can be For a beam with a given width, the hyperelastic beam model can
selected in order to maximize the responsivity. As the dependencies predict the deformation when the beam is subjected to these forces,
4
J. Oprel et al. Additive Manufacturing 73 (2023) 103674
Table 1
Parameters for the design, resulting from the optimization.
Parameter Description Value large Value small
design design
𝜇 Shear modulus 132 MPa 132 MPa
(Armadillo)a (Armadillo)a
(𝜇)2 Shear modulus in design for 4 MPa
experiments (Ninjaflex)a
𝐷max Design error due to normal 20% 20%
force
𝑆b Beam length 15 mm 5 mm
𝐻 Beam thickness 2.81 mm 0.98 mm
𝑊tot Total width 22.22 mm 8.75 mm
𝑊open Width open space 0.90 mm 0.32 mm
Fig. 7. Beam width to length (𝐻∕𝑆𝑏 ) ratio as a function of the width of the electrical 𝑊elect. Width electric layout 14.7 mm 6.2 mm
structure with the acceptable relative deviation in the horizontal displacement (due to 𝐿 Length of wires, width of 22 mm 8.75 mm
normal stresses) as parameter. beams in 𝑧 direction
𝑆𝑦, in groups Horizontal wire separation in 0.6 mm 0.6 mm
groups
with and without normal stress. Using a numerical solver in Matlab, 𝑆𝑦, groups Separation between groups 4 mm 3 mm
𝑆𝑥 Vertical separation 1.2 mm 0.7 mm
based upon the Trust-Region-Dogleg Algorithm (fsolve()), an in-
𝑑 Wire diameter 0.4 mm 0.4 mm
verse problem can be solved: finding the beam width for which the 𝑛groups Number of groups 3 2
difference between the situation with and without normal stress is a 𝑛wires Number of wires per group 3 2
given an allowed deviation 𝐷max defined as: 𝑦0 Initial horizontal displacement 0.8 mm 1.35 mm
of wires
𝛥𝑦|𝑃 =0,𝑄=𝑄max − 𝛥𝑦|𝑃 =𝑃max ,𝑄=𝑄max
𝐷max ∶= (2) a
Value based on the Young’s modulus from the manufacturers datasheets [36,37] and
𝛥𝑦|𝑃 =0,𝑄=𝑄max a Poisson ratio of 0.5 determined for generic TPUs [38].
where, 𝛥𝑦 is the deflection of the beam in the 𝑦 direction (for different
load cases). This is an estimate for the maximum deviation that can
be expected in the capacitance difference that is measured due to the
normal stress. However, the deviation of the capacitance change is also
influenced by the electrodes moving closer together in the 𝑥 direction.
Nevertheless, the modeling results indicate that the horizontal devia-
tion is the dominant cause for the deviation because the capacitance
is measured differentially. By solving the described inverse problem,
Fig. 7 is obtained. Notice that a higher accepted deviation results in
a smaller beam width, as more flexibility is accepted. The results from
Fig. 7 are used to obtain the appropriate beam width after the electrical
layout has been determined.
5
J. Oprel et al. Additive Manufacturing 73 (2023) 103674
6
J. Oprel et al. Additive Manufacturing 73 (2023) 103674
factor (tan 𝛿) of the flexible Ninjaflex material. The DMA experiment experimental results are also quite large. This could be caused by
is executed with similar strain levels and at the same frequency of the anisotropic properties of printed materials and inaccuracies in the
0.5 Hz as the sine wave input used for the experiments with the printed structures, for example additional material deposition caused
sensor. The damping factor that resulted from the DMA analysis was by oozing which can be seen in Fig. 11. The models are thus not suitable
approximately 0.6. For the same sine wave input, the experiments with to predict measurement outcomes precisely, but the models can be used
the sensor show a damping factor that is on average twice as high. For for optimization of the sensor design.
a detailed description of the methods and results used in this analysis, The two materials have very similar shear ( modulus
) normalized
see Appendix D. The cause of the mechanical hysteresis of the sensor responsivities in the total sensor behavior 𝜇 d𝐶 . Because the Ar-
can thus only partially be explained by the internal hysteresis of the d𝜏
madillo material has a higher Young’s modulus, the measurement range
material. Additional losses are potentially caused at the interfaces, such
of this sensor is larger, whilst the (non-normalized) responsivity is
as the flexibility in the connection of the force actuator or the clamping
reduced by the same factor. For the sensors printed with the stiffer
in the vise.
Armadillo material, the mechanical behavior shows a substantially
For the results in Figs. 14 and 15, a linear least squares fit is made.
more flexible behavior than expected, whilst the electrostatic behavior
The slope of this fit can be compared to the modeling results. The re-
seems substantially less sensitive. These values should be similar to
sults are shown in Table 2. There are substantial discrepancies between
the analytical and the FE models. For the mechanical model, this can the values for the sensors printed with the Ninjaflex material. Because
be explained by the fact that the FE model simulates the flexibility the total behavior is as expected, it is likely that the deviations are
of the complete 3D structure of the sensor, whilst the beam model caused by some flexibility in the measurement setup. It is expected
assumes that only the elements on the sides of the sensor are flexible. that this only influences the separate results with the stiffer material,
The extra flexibility in the top and bottom structure of the sensor is as the forces are a factor of 6 larger than in the measurements with
not taken into account in the beam models, which leads to smaller the flexible Ninjaflex material. The responsivity of the large sensors
expected deviations. The difference between electrostatic models is (Armadillo) is 2.3 pF∕MPa. The responsivity of the small sensor design
partially caused by the more plate like geometry of the electrodes. 8.75 × 8.75 × 11 mm is 0.9 pF∕MPa.
The difference between the different electrostatic models is shown in Compared to the sensor by Mertodikromo et al. [30] which is
more detail in Appendix A.4. The deviations between the models and designed for the same application and measurement range, the main
7
J. Oprel et al. Additive Manufacturing 73 (2023) 103674
Fig. 12. Sensor positioning and parts of the measurement setup for sensor characterization.
advantages of the smallest sensor reported in this work is a lower sur- For the analysis of the bandwidth of the sensor some additional
face area (0.77 cm2 compared to 3 cm2 ) and a factor 5 improvement in measurements were done with a square-wave signal. The bandwidth of
the responsivity (0.9 pF∕MPa compared to 0.18 pF∕MPa). Furthermore, the mechanical behavior is at least 10 Hz as required for the application
the sensor is fully printed, whilst the conductive parts are not printed in shoe insoles. The capacitance was read out at only 8 Hz in this work,
in the sensor developed by Mertodikromo et al. The sensor presented in but the sample frequency can be increased using different settings in
this work is limited to capture the shear force in one direction, whilst the hardware at the expense of a lower signal to noise ratio. As for the
the sensor reported by Mertodikromo et al. can measure in both shear electrical dynamics these are not expected to limit the bandwidth as
dimensions. Also, the sensor reported here is thicker (11 mm compared the excitation frequency is at 16 kHz [40].
to 3 mm). This could be improved in a later iteration, at the cost of These first results are promising: the printed sensors seem to be
responsivity. applicable as shear stress sensor for measurement on the human foot.
8
J. Oprel et al. Additive Manufacturing 73 (2023) 103674
The design can relatively easily be adjusted to other requirements: the Table 2
Fitted slope of the measurement results for the large and small sensor design made
responsivity can be improved by reducing the measurement range or
from Ninjaflex and Armadillo, compared to the model predictions (only for the large
vice versa. This can be done by modifying the width of the beams in the sensor design). The two numbers correspond to two input signals: sine and trapezium.
design, or by using a different material. E.g. in this research Ninjaflex The letters (a, b) are used to distinguish two printed sensors that ideally would be
is used in one of the designs to reduce the force measurement range of identical. A measurement error has likely influenced the measurements for the sensor
printed with the stiffer Armadillo TPU, these measurement results are put in brackets.
the sensor by a factor of 33 and improve the responsivity by the same
The small sensor design cannot be compared to the model prediction shown here.
factor. Also, one design is smaller, at the cost of having a decreased ( ) ( )
Sensor Total 𝜇 d𝐶 Mechanic
( ) Electrostatic d𝐶
responsivity. With (one of) these slight modifications, the sensor design d𝜏
d𝑦
𝜇 d𝜏 in mm
d𝑦
in pF in pF/mm
could be made suitable for shear force measurements in a variety
Model prediction 344 478 0.72
of other applications. Further, future research into the application of (approximation)
model based hysteresis compensation may improve the accuracy of the Model prediction 711 773 0.92 (finite length,
measurements [41]. Improving the accuracy of the printing process (FE model) rounded plate)
Large design 309–363 451–585 0.69–0.61
may improve repeatability and accuracy of the measurements. Addi-
(Ninjaflex) a
tionally, further research is needed to solve the challenge of interfacing Large design 243–261 374–396 0.66–0.67
larger sensing structures, consisting of multiple sensors. (Ninjaflex) b
Large design 247–275 (1525–1572) (0.16–0.18)
(Armadillo) a
Large design 268–304 (1372–1435) (0.20–0.21)
(Armadillo) b
Small design 119
(Armadillo)
5. Conclusion
9
J. Oprel et al. Additive Manufacturing 73 (2023) 103674
Declaration of competing interest individual points on the surface, as this would make infinitely large
potentials [46]. Furthermore, calculations are simplified substantially
The authors declare that they have no known competing financial by assuming a homogeneous charge distribution: For the fields outside
interests or personal relationships that could have appeared to the wire, this is equivalent to having a line charge in the center of the
influence the work reported in this paper. wire. The effect of this approximation is that the surface of the wires are
not necessarily equipotentials, which is non-physical for conductors. In
Data availability Appendix A.3 it will be shown that the effect of this approximation
is small if the wires have a large separation. However, there is not
Models and optimization tools are made available [32] one single value for the potential difference between two wires: this
depends on which points on the wires are used to define the potential.
Acknowledgments To reduce the dependency on the choice of these match-points, the
average of multiple (𝑛p ) points is used (see Fig. 3). With this, it is
The authors would like to thank ir. Dimitris Kosmas for helping with possible to analytically express the contribution of a wire (𝑖) on the
the capacitance measurements using the AD7747 and ing. Andries van potential difference between two wires (𝑗 and 𝑘). This can be done
Swaaij from the Elastomer Technology and Engineering group (Univer- easily by integrating the electric field along the orange dashed path
sity of Twente) for facilitation and assistance with DMA. This work was in Fig. A.16: only radially outwards and along an equipotential (shown
financially supported by the 4TU Dutch Soft Robotics Consortium. in light gray thin lines):
| |
𝜆𝑖 1 ∑ ⎛⎜ ||𝑟⃗𝑖,𝑗,𝑚 || ⎞⎟
𝑛p
Appendix A. Derivation and discussion on the center charge ap-
𝛥𝜙̃(𝑗,𝑘),𝑖 = ln (A.1)
proximation for the capacitance of conductive wire structures 2𝜋𝜖0 𝑛p 𝑚=1 ⎜ ||𝑟⃗𝑖,𝑘,𝑚 || ⎟
⎝ ⎠
In this appendix, the center charge approximation is derived and Here 𝜆𝑖 is the line charge on wire 𝑖. 𝑟⃗𝑖,𝑗,𝑚 denotes the distance from
discussed. This method provides a quick way to approximate the capac- wire 𝑖 to the 𝑚-th match-point on wire 𝑗 as in Fig. A.16 and 𝜖0 is the
itance of conductive wire structures that is useful to optimize sensing permittivity. The term in the sum (that is being averaged) is the natural
structures. In Appendix A.1, the approximation method is derived. In logarithm of the distance to the point on wire 𝑗 minus the natural
Appendices A.2 and A.3 the method is compared to similar numerical logarithm of the distance to the point on wire 𝑘. This is proportional to
methods and an analytical result. the potential difference between these points caused by charge on wire
𝑖.
A.1. Derivation of the center charge approximation From Eq. (A.1), it can be noted that the contribution of wire 𝑖 to
the potential difference is linear in the line charge on wire 𝑖 (𝜆𝑖 ). As
To find the capacitance of conductive wire structures, the charge wire 𝑖 is an arbitrary wire, this also holds for any other wire. Adding
distribution should be found in a situation where there is a given these contributions yields an expression for the potential difference
potential difference between the wires. Analytic solutions to this prob-
which is linear in all (𝑛) line charges. For 𝑛 wires, 𝑛 − 1 independent
lem only exist for simple structures. In more general cases, numerical
linear equations for the potential difference can be made in the same
techniques are needed to arrive at an estimate for the charge distri-
manner. By requiring a net charge of 0, it is ensured that the potential
bution. One way to achieve this is to make an assumption on the
infinitely far away is 0 (even in this model where the wires are
charge distribution on the wires. Such approaches have been studied
infinitely long). In this way, an invertible linear system of equations is
before [42–45]. The differences with these method will be elaborated
obtained. This allows to calculate the charge on each wire for a given
at the end of this subsection.
potential difference. By summation of these charges, the capacitance
In the approximation technique used in this work, it is assumed that
between two wire structures can be determined from the definition
the charge on the wire is homogeneously distributed on the surface of
of capacitance (𝐶 = 𝑉𝑄 ). If it is assumed that the potential of the
the wires. This is the simplest assumption one can make for the charge
two (bottom) electrodes for differential measurement is equal (which
distribution: It is impossible to assume that the charge is situated at
it is with the measurement equipment used in this research [40]), this
relation can also be used for a case with 3 electrodes: Because the
potential difference is the same, the capacitance will be proportional
to the charge on each structure.
10
J. Oprel et al. Additive Manufacturing 73 (2023) 103674
approximation is compared to an analytical expression for the capac- Appendix B. Timoshenko beam model for the required boundary
𝜋𝜖0
itance per unit length, which is given by 𝐶𝐿 = −1 [47]. The conditions
cosh (𝑆∕𝑑)
approximation error depends on the separation between the wires. In
To model the beam structures of the sensor design, the beam model
Fig. A.17, the approximation error is shown for a range of separations
developed by He et al. [35] is used to predict the deformation of a beam
and different numbers of match-points. For a low number of points
which is fixed on the bottom and is loaded on its end by a normal force
(1 or 2), there is a large variation in error values. This is because
for each number of match-points, all combinations of equally spaced 𝑃 and a shear force 𝑄, see Fig. B.19. This is a model of a Timoshenko
positions on the surface of the wires are studied. This results in a beam with an incompressible Neo-Hookean material. Deformation is
range of possible deviations. It is clear that the uncertainty due to this modeled in two dimensions: deformations in the out of plane direction
positioning reduces quickly as more match-points are used. Also, the in Fig. 1 are not considered. Modeling a Timoshenko beam means
maximum error decreases as the number of match point increases. For that the angle of the normal of the deformed cross-sections 𝜑 is not
𝑛𝑝 ≥ 4, the error in the approximation is 5% or lower if the (center-to- necessarily equal to the angle of the mid-plane of the beam 𝜃 (see
center) separation over diameter ratio (𝑆∕𝑑) is larger than 2. This is in Fig. B.19(b)). The difference between these angles is the shear angle
line with the result found in Section 2.2.1, because a separation of 2𝑑 𝛼. Because this shear is taken into account in the model, the model
corresponds to a relative wire density of 50%. is suitable to apply for relatively thick beams. The deformation of the
cross-section is taken in account to the extent that the cross-sections
A.4. Comparison with 2D and 3D FE models are still planar, but the thickness of the beam can vary depending on
the stresses on the material. With this the incompressibility is modeled.
The modeling result of the center charge and FE methods are com- This thickness stretching effect is only taken into account for one
pared for a structure as in Fig. 2. Three FE models are used: The first is direction. The length of the beam in the 𝑧-direction of Fig. B.19 is
assumed to remain constant.
With these assumptions, He et al. [35] derived the following expres-
sions for the equilibrium of beams:
( ( ) )′
1 (𝜑′ )2 𝐻 2 𝑝
𝛬 cos 𝜃 − + 7 cos 𝜑 + =0 (B.1a)
𝛬3 cos3 𝛼 𝛬 cos7 𝛼 𝜇𝐴c
Fig. A.18. Comparison of the center-charge approximation with FE models for a wire Fig. B.19. Sketches of a deformed beam with fixed rotation and external forces at
structure as in Fig. 2. 𝑥 = 𝑆b , with relevant parameters and loads.
11
J. Oprel et al. Additive Manufacturing 73 (2023) 103674
( ( ) )′
1 (𝜑′ )2 𝐻 2 𝑞
𝛬 sin 𝜃 − + 7 sin 𝜑 + =0 (B.1b)
𝛬3 cos3 𝛼 𝛬 cos7 𝛼 𝜇𝐴c
( ) ′ ( )
1 (𝜑′ )2 𝐻 1 (𝜑′ )2 𝐻 2 𝑚
− tan 𝛼 + + =0 (B.1c)
3 𝛬6 cos6 𝛼 𝛬2 cos2 𝛼 𝛬6 cos6 𝛼 𝜇𝐴c
A prime (′ ) denotes a derivative with respect to 𝑥. 𝑥 is not explicitly
part of Eq. (B.1), but 𝛬, 𝜃, 𝜑, 𝛼, 𝑝, 𝑞 and 𝑚 can depend on 𝑥. 𝑝,
𝑞 and 𝑚 denote the distributed horizontal force, vertical force and
bending moment per unit length respectively, and 𝛬 is the stretch of
the mid-plane element (the length of the deformed element divided by
its original length).
In the cited work by He et al. the model is derived and example
applications are presented for some load cases. However, the load case
needed to model the sensing structure is not included. In this appendix
Fig. D.20. Mechanical behavior of the large Ninjaflex sensor a (of the fifth and last
the boundary conditions for this load case are applied: It is assumed iteration): force and position over time for the last two periods of a 0.5 Hz sine wave
that the bottom of the beam is clamped and the top of the beam remains input. A scaled version of the output is fitted to the input to estimate the damping
horizontal, that is: factor.
• At 𝑥 = 0, 𝑢0 = 𝑣0 = 𝜑 = 0.
• At 𝑥 = 𝑆b , the applied force in negative 𝑥 and 𝑦 directions are 𝑃 D.1. Analysis of delay in the sensor measurements
and 𝑄 respectively and 𝜑 = 0
With these boundary conditions, the equations for equilibrium (B.1) can The delay between force and position is determined for measure-
be rewritten to: ments with a sine force input of 0.5 Hz. This is done for 7 sensors
( ) printed with Ninjaflex TPU. This includes the two sensors of the fi-
1 (𝜑′ )2 𝐻 2 𝑃 cos 𝜑 + 𝑄 sin 𝜑
𝛬 cos 𝛼 − + 7 =− (B.2a) nalized design, of which results are reported in Section 4, but also
𝛬3 cos3 𝛼 𝛬 cos7 𝛼 𝜇𝐴c
5 sensors from previous iterations. In these iterations the mechanical
𝑃 sin 𝜑 − 𝑄 cos 𝜑
𝛬 sin (𝛼) = (B.2b) design is identical, only the connections to the capacitive layout are
𝜇𝐴c slightly different between iterations.
( ) ′
1 (𝜑′ )2 𝐻 𝑃 sin 𝜃 − 𝑄 cos 𝜃 To determine the delay in these measurements, the input 𝐹 (force)
=𝛬 (B.2c)
6
3 𝛬 cos 𝛼 6 𝜇𝐴c is compared to a scaled and shifted output 𝑥 (position):
̃ = 𝑎𝑥(𝑡 − 𝑑) + 𝑏
𝑥(𝑡) (D.1)
The remaining boundary conditions for this reduced problem are
𝜑(0) = 𝜑(𝑆b ) = 0. A numerical approach is used to solve this sys- Here, 𝑎, 𝑏, and 𝑑 are parameters. An estimate for these parame-
tem of equations, similar to the numerical approach described by ∑
̃ 2 using a simplex method
ters is found by minimizing 𝑡 (𝐹 (𝑡) − 𝑥(𝑡))
He et al. [35]. The implementation of this model in Matlab is made (fminsolve() in Matlab). An example of such a fit is shown in
available online [32]. Fig. D.20.
The delay is used to calculate the damping factor tan(𝛿) that one
Appendix C. Slicing and printing settings expects based on these measurements (assuming all damping can be
attributed to the material properties), this can be compared to the value
The settings for printing of the different materials as configured in found from DMA, see Appendix D.2. The phase shift (𝛿) is found by
the slicer (Simplify3D) can be found in Table C.3. The cooling is turned multiplying the delay with the angular frequency, 𝜋 in case of the
off for the first layer and for the layers at which the support material 0.5 Hz sine wave input. The value of tan(𝛿) for the 7 sensors is shown
is first printed to improve adhesion. in Table D.4.
Table C.3
Printing settings as configured in the slicer.
D.2. Dynamic mechanical analysis (DMA) of Ninjaflex TPU
Ninjaflex/Armadillo PI-ETPU (conductive) BVOH (support)
In dynamic mechanical analysis (DMA), the goal is to determine
Nozzle number 2 3 5
Nozzle temperature 235 °C 225 °C 210 °C the damping factor (tan(𝛿)) of the flexible material (Ninjaflex). For
Nozzle diameter 0.4 mm 0.4 mm 0.4 mm the DMA, a sample of 2 × 6 × 50 mm is printed flat on the build-
Extrusion multiplier 1.3 1.15 0.75 plate with the same printer settings as the sensor (with 3 wall layers
Coast at end 1 mm 0 0 and a diagonal infill of 100%). The dynamic behavior of materials
Bed temperature 60 °C is potentially dependent on the loading: especially the frequency of
Layer height 0.2 mm
the loading, dynamic strain (magnitude) and static strain (offset) can
Cooling 0% for layers 1, 15 and 93 100% for other layers
influence the behavior. Therefore, these parameters should be chosen
such that the DMA experiment is similar, or at least relevant, to the
Appendix D. Characterization of the mechanical behavior of Nin- characterization of the sensor. For this reason, the experiments are
jaflex TPU using DMA executed with a frequency of 0.5 Hz. From analytic estimations and FE
simulations, it is expected that the maximum strain that occurs in the
In this Appendix, it is investigated to what extent the material sensor is around 4%. Therefore, DMA experiments are executed with
properties of Ninjaflex TPU can explain the hysteresis that is present in 3, 4 and 5% dynamic strain. In the main experiments with the sensor
the mechanical behavior of the printed sensors, see Section 4, Fig. 14. the excitations have a relatively small offset: shear forces are applied
In Appendix D.1, the delay in the sensor response is analyzed. From in two directions with little static strain. This is not possible in DMA:
the delay between force and position, the phase shift 𝛿 is found. In Ap- the nature of these experiments requires that the static strain is larger
pendix D.2, it is described how a dynamic mechanical analysis (DMA) than the dynamic strain such that there is no buckling. Multiple static
was performed to find the damping factor tan(𝛿) of printed Ninjaflex strains are analyzed, such that the results can be used to extrapolate to
TPU. The results are summarized and discussed in Appendix D.3. zero static strain. Measurements are performed using a Netzsch Gabo
12
J. Oprel et al. Additive Manufacturing 73 (2023) 103674
13
J. Oprel et al. Additive Manufacturing 73 (2023) 103674
[25] X. Aeby, R.v. Dommelen, D. Briand, Fully FDM 3D printed flexible capacitive [36] Armadillo 3D Printing Fillament: Semi-rigid Polyurethane Material for FDM
and resistive transducers, in: 2019 20th International Conference on Solid- printers (Technical Specifications), NinjaTek, Manheim, USA, 2017, Available
State Sensors, Actuators and Microsystems & Eurosensors XXXIII (TRANSDUCERS online: https://ninjatek.com/armadillo/ [Accessed: 10-03-2021].
& EUROSENSORS XXXIII), 2019, pp. 2440–2443, http://dx.doi.org/10.1109/ [37] NinjaFlex 3D Printing Fillament: Flexible Polyurethane Material for FDM printers
TRANSDUCERS.2019.8808268. (Technical Specifications), NinjaTek, Manheim, USA, 2016, Available online:
[26] L.Y.W. Loh, U. Gupta, Y. Wang, C.C. Foo, J. Zhu, W.F. Lu, 3D printed https://ninjatek.com/ninjaflex/ [Accessed: 23-02-2021].
metamaterial capacitive sensing array for universal jamming gripper and human [38] H.J. Qi, M.C. Boyce, Stress–strain behavior of thermoplastic polyurethanes, Mech.
joint wearables, Adv. Eng. Mater. 23 (5) (2021) 2001082, http://dx.doi.org/10. Mater. 37 (8) (2005) 817–839.
1002/adem.202001082. [39] G. Wolterink, A. Umrani, M. Schouten, R. Sanders, G. Krijnen, 3D-printed
[27] M. Schouten, C. Spaan, D. Kosmas, R. Sanders, G. Krijnen, 3D printed capacitive calorimetric flow sensor, in: 2020 IEEE SENSORS, 2020, pp. 1–4, http://dx.doi.
shear and normal force sensor using a highly flexible dielectric, in: 2021 IEEE org/10.1109/SENSORS47125.2020.9278640.
Sensors Applications Symposium (SAS), 2021, pp. 1–6, http://dx.doi.org/10. [40] 24-Bit Capacitance-to-Digital Converter with Temperature Sensor: AD7747, Ana-
1109/SAS51076.2021.9530032. log Devices, Norwood USA, 2007, Available online: https://www.analog.com/
[28] G. Wolterink, R. Sanders, B.-J. van Beijnum, P. Veltink, G. Krijnen, A 3D- en/products/ad7747.html#product-overview [Accessed: 15-4-2020].
printed soft fingertip sensor for providing information about normal and shear [41] D. Kosmas, M. Schouten, G. Krijnen, Hysteresis compensation of 3D printed sen-
components of interaction forces, Sensors 21 (13) (2021) 4271, http://dx.doi. sors by a power law model with reduced parameters, in: 2020 IEEE International
org/10.3390/s21134271. Conference on Flexible and Printable Sensors and Systems (FLEPS), 2020, pp.
[29] K. Kim, J. Park, J. hoon Suh, M. Kim, Y. Jeong, I. Park, 3D printing of multiaxial 1–4, http://dx.doi.org/10.1109/FLEPS49123.2020.9239580.
force sensors using carbon nanotube (CNT)/thermoplastic polyurethane (TPU) [42] C.R. Paul, Analysis of Multiconductor Transmission Lines, Wiley-IEEE Press,
filaments, Sensors Actuators A 263 (2017) 493–500, http://dx.doi.org/10.1016/ 2008, pp. 160–239, http://dx.doi.org/10.1109/9780470547212.ch5, Ch. The
j.sna.2017.07.020. PerUnitLength Parameters for Multiconductor Lines.
[30] J. Mertodikromo, F. Zorin, C.J. Lee, A low-profile shear force sensor for wearable [43] J.C. Clements, C.R. Paul, A.T. Adams, Computation of the capacitance matrix
applications, IEEE Sens. J. 20 (18) (2020) 10453–10459, http://dx.doi.org/10. for systems of dielectric-coated cylindrical conductors, IEEE Trans. Electromagn.
1109/JSEN.2020.2985396. Compat. EMC-17 (4) (1975) 238–248, http://dx.doi.org/10.1109/TEMC.1975.
[31] A. Dijkshoorn, M. Schouten, G. Wolterink, R. Sanders, S. Stramigioli, G. Krijnen, 303430.
Characterizing the electrical properties of anisotropic, 3D-printed conductive [44] J. Savage, W. Smith, C. Paul, Moment method calculation of the per-unit-
sheets for sensor applications, IEEE Sens. J. 20 (23) (2020) 14218–14227, length parameters of cable bundles, in: Proceedings of IEEE Symposium on
http://dx.doi.org/10.1109/JSEN.2020.3007249. Electromagnetic Compatibility, 1994, pp. 441–446, http://dx.doi.org/10.1109/
[32] J. Oprel, G. Krijnen, G. Wolterink, J. Schilder, Models and optimization tools ISEMC.1994.385608.
for a novel 3d printed capacitive shear stress sensor, 4TU.ResearchData (2023) [45] M. Saih, H. Rouijaa, A. Ghammaz, Computation of multiconductor transmission
http://dx.doi.org/10.4121/19188098. line capacitance using method of moment, in: Proceedings of 2014 Mediterranean
[33] A. Bower, Applied Mechanics of Solids, CRC Press, 2009. Microwave Symposium (MMS2014), 2014, pp. 1–6, http://dx.doi.org/10.1109/
[34] T. Reppel, K. Weinberg, Experimental determination of elastic and rupture MMS.2014.7088781.
properties of printed Ninjaflex, Tech. Mech.-Eur. J. Eng. Mech. 38 (1) (2018) [46] M.A. Gottlieb, R. Pfeiffer, The Feynmann Lectures on Physics, Vol. II, California
104–112, http://dx.doi.org/10.24352/UB.OVGU-2018-010. Institute of Technology, 1965, Ch. 5 Application of Gauss Law..
[35] L. He, J. Lou, Y. Dong, S. Kitipornchai, J. Yang, A shearable and thickness [47] H.E. Green, A simplified derivation of the capacitance of a two-wire transmission
stretchable finite strain beam model for soft structures, Meccanica 53 (15) (2018) line, IEEE Trans. Microw. Theory Tech. 47 (3) (1999) 365–366, http://dx.doi.
3759–3777, http://dx.doi.org/10.1007/s11012-018-0905-4. org/10.1109/22.750243.
14