Available online at www.sciencedirect.

com

ScienceDirect
Materials Today: Proceedings 12 (2019) 252–261 www.materialstoday.com/proceedings

DAS35

3D-printed ankle-foot orthosis: a design method

Alberto Dal Maso, Francesca Cosmi*
University of Trieste, Via A. Valerio, 1, 34100, Trieste, Italy.

Abstract

Additive manufacturing offers several advantages when compared to conventional manufacturing technologies, especially where
large scale customization is requested. In this article, a procedure for the design of a fully-customized 3D-printed ankle-foot
orthosis (AFO) is described, with reference to a case study regarding a 21-year-old woman with an injured ankle. The first step is
the acquisition of the geometrical data from the patient’s foot: this is done using photogrammetry. The second step is the data
conversion and import in a CAD modeler; SolidWorks™ is chosen for this purpose. The AFO is modelled parametrically around
the foot mesh and optimized in order to resist the predicted mechanical stresses. Finally, the device is 3D-printed on an FDM
printer and tested on the patient. An excellent geometrical correspondence between the AFO and the patient’s foot is highlighted:
this leads to great comfort and enhances medical functionality. The described procedure can be easily automated, further
reducing the lead time and costs of the whole process.
© 2019 Elsevier Ltd. All rights reserved.
Selection and peer-review under responsibility of 35th Danubia Adria Symposium on Advances in Experimental Mechanics.

Keywords: 3D printing; Ankle Foot Orthosis; photogrammetry; FEM simulation; FDM.

1. Introduction

Additive manufacturing offers several advantages when compared to conventional manufacturing technologies,
especially where large scale customization is requested [1]. This is particularly common for certain medical
applications such as orthopedics, in which the efficiency of a treatment is strongly connected with each individual
patient’s anatomical geometry [2].
An Ankle-Foot Orthosis (AFO) is an external medical device which helps a patient hold his/her ankle in the right
position by providing external mechanical support when the muscular/skeletal system alone is not sufficient. Fig. 1
shows some examples of different kinds of AFOs [3].

* Corresponding author. Tel.: +39-040-558-3431; fax: +39-040-558-3812

E-mail address: cosmi@units.it

2214-7853 © 2019 Elsevier Ltd. All rights reserved.

Selection and peer-review under responsibility of 35th Danubia Adria Symposium on Advances in Experimental Mechanics.
 A. Dal Maso, F. Cosmi / Materials Today: Proceedings 12 (2019) 252–261 253

This research is based on a case study concerning a 21-year-old woman who had difficulties in walking after
wearing a plaster cast for three weeks; an AFO can help her post-traumatic rehabilitation. The more the shape of the
orthosis adapts to the patient’s body, the more comfortable and efficient the treatment will be. This paper describes a
method to design a fully-customized Ankle-Foot Orthosis, taking advantage of 3D-scanning and 3D-printing
techniques.

Fig. 1: Some examples of ordinary Ankle-Foot Orthoses.

2. Materials and methodology

The adopted design process can be divided into the following steps.

2.1. Foot scan with photogrammetry

The first main objective is to design the internal geometry so that it corresponds perfectly to the anatomical shape
of the patient’s foot. Therefore, the geometrical data of the area around the ankle has to be acquired. One possible
way of obtaining this data without the use of cumbersome and expensive 3D scanners is photogrammetry [4].
Photogrammetry is an extremely versatile technique that allows the generation of a 3-dimensional CAD model of
virtually any object from a series of pictures (Fig. 2). These pictures must be taken from different angles and should
include as much information as possible regarding the examined subject. Key to a good reconstruction is high
quality of the images: sharpness and uniform lighting are fundamental. The background should be either single-color
or, preferably, should contain recognizable features or color patterns which help during the alignment phase. The
object itself should be rigid and must hold perfectly still; moreover, it should not have any transparent or shiny
surfaces. It is preferable to avoid wide uniformly-colored areas that do not present valuable recognition patterns. In
theory, each point of the object should appear in at least three pictures in order to determine its exact position. In
practice, it is advisable to be far more redundant: some hundreds of pictures taken from all angles are generally
enough to guarantee a precise result.
With reference to the aforementioned characteristics, we notice that photogrammetry is suitable for 3D
reconstruction of anatomical parts, as long as these can hold still during the whole image acquisition process. In this
case, photogrammetry is applied to the patient’s leg. To ensure that the part remains in the same position throughout
the process, a comfortable position for the patient must be found, for example on a tall stool.
150 photos of the patient’s foot were taken from different angles and were then loaded into the Agisoft
PhotoScan Pro™ software. PhotoScan™ guides the user through the necessary steps in order to obtain the 3D
model, however some additional steps were necessary to ensure the highest possible quality of results [5]. The
adopted procedure is described in the following points.
 Photo alignment. In this phase, the software reconstructs the position of the cameras for each shot and estimates
the position of tie points, thus obtaining a tie point cloud. Tie points are points that are recognized by the
software as common to two or more images: the geometry of the object and the position of cameras is calculated
based on tie points.
254 A. Dal Maso, F. Cosmi / Materials Today: Proceedings 12 (2019) 252–261

 Tie point cloud and alignment optimization. This step is not directly part of the workflow, but it is of fundamental
importance to obtain good overall results. Using the gradual selection tool, all points with a high value of
reconstruction uncertainty are deleted. The same is done for the reprojection error and for the projection
accuracy.
 Sparse cloud cleaning. All points belonging to the background or any other undesired detail must be manually
erased. This can be easily done with the lasso tool provided by the software.
 Dense point cloud creation. At this stage, Photoscan™ creates a dense point cloud from the available images.
This step is the most computationally demanding and might take up to 80% of the total calculation time of the
entire photogrammetric reconstruction.
 Dense point cloud cleaning. Once the dense point cloud is obtained, it has to be further cleaned from any
unwanted points, be them part of the background or just evidently out of place. If the reconstruction went well,
this phase should not take much effort.
 Mesh creation. In this final step, the software builds a 3D mesh from the dense point cloud. Photoscan™ has a
very efficient algorithm for this task, therefore it generally takes just seconds to be completed.
 Mesh exportation, directly in STL format. This will very likely produce an excessively big file, which may be
difficult to manipulate. Therefore, it is advisable to reduce the face count of the mesh through the decimate mesh
tool. A target face count of around 20 000 is a good compromise between detail preservation and data
manageability.

Fig. 2: Some photos of the foot and the resulting 3D model.

Table 1 summarizes the main parameters for the point clouds and the mesh that were calculated in our case
study. All calculations were performed on an Apple™ Macintosh™ machine with an Intel™ Core i7 3.6 GHz quad
core processor and 16 GB of RAM.
 A. Dal Maso, F. Cosmi / Materials Today: Proceedings 12 (2019) 252–261 255

Table 1. Properties of the point clouds and the mesh calculated for the case study.
Count Calculation time
Tie point cloud 26 645 points 2h 07min
Sparse point cloud 1 523 points -
Dense point cloud 512 358 points 6h 34min
Full mesh 74 384 faces 0h 2min
Decimated mesh 20 000 faces -

2.2. Data import in CAD modeler

The second step is importing the foot mesh into a CAD modeler. This mesh will serve as a reference for the AFO
design. There are many programs that can handle large STL files, however these usually do not allow a good
parametrization of the project. The mechanical CAD software SolidWorks™ was chosen for this purpose. Instead of
struggling when dealing with large meshes, SolidWorks™ can generate very well-parametrized models, which
becomes fundamental during the upcoming topological optimization phase.
Direct editing of large STL (or equivalent mesh files) in SolidWorks™ is very complex, if not virtually
impossible, due to the numerous inevitable self-intersecting faces which often occur during the meshing process.
Thanks to the Scan to 3D add-in, the program manages to fix most errors automatically, however, in order to
perform any modelling around it, SolidWorks™ must recognize the imported object as a surface or solid body. In
most cases this is done automatically, but for complex geometries such as anatomical shapes the process is more
elaborated. We chose to draw a loft around the foot mesh; the loft feature creates a shape by making transitions
between multiple planar profiles along an approximate path using polynomial surfaces.
The steps are the following.
 Mesh import. The STL mesh was imported in a new SolidWorks™ part.
 Curve generation. With the help of the Curve Wizard, a series of cross-section sketches was generated. In total,
70 equally-spaced cross-sections were produced along the leg and foot.
 Loft Generation. A loft passing through all these sections was created. This is a delicate operation and its success
depends very much on how regularly spaced the cross-sections are. In this case, the use of guide curves did not
improve the result, therefore no guide curves were used. Once the loft has been correctly created, a SolidWorks™
feature is produced and can now be manipulated in this software. It is advisable to manually double-check
geometrical correspondence between the original mesh and the loft.
 Finally, it is necessary to scale the mesh in order to have an exact correspondence between the lengths in the
model and in reality. To do so, at least one distance measure ( ) must be taken on the real foot. This will be
compared with the virtual measure of the same distance on the model ( ). The model will have to be scaled
by a factor so that:

2.3. AFO modelling

The orthosis modelling is divided in three main parts: inner surface modelling, outer surface modelling, and
shaping.
A certain space must be left between the inner surface and the foot, in order to compensate for eventual
geometrical errors and to allow some space for the foot if this is swollen (Fig. 3). The inner surface is generated
through the offset tool. The size of the offset is not universally agreed upon and must be determined by experience.
256 A. Dal Maso, F. Cosmi / Materials Today: Proceedings 12 (2019) 252–261

Fig. 3: Generation of the inner surface through the offset feature.

The outer surface is generated with a boundary boss/base feature in SolidWorks™ (Fig. 4). This feature is similar
to the loft, but allows somewhat more control over the resulting geometry. The procedure involves creating some
section sketches that will be connected to form the desired surface. Opposed to the foot-loft creation, there is no
exact geometry that has to be traced precisely, therefore fewer sections are required: this allows greater stability and
efficiency of the program while, on the other hand, sacrificing part of the control on the resulting geometry. The
following steps are adopted to create the outer surface:
 A series of 16 equally-spaced planes are drawn along the foot.
 Using the Curve Wizard, the intersection of the inner surface with each plane is generated.
 Thanks to the sketch-offset tool, a new profile is generated on each plane from the corresponding intersection
sketch. The entity of each offset is an important parameter, which can vary in relation to each section.
 Once the boundary boss/base is performed, a new solid body is obtained. The inner surface is now subtracted
from this through the Boolean operator; in this way, a shell with variable thickness is created.

Fig. 4: Generation of the outer surface through the boundary boss/base feature.

On principle, this device would serve its purpose, but it would be impossible to wear in practice. Therefore, an
opening must be designed, to allow the foot to be comfortably inserted and to guarantee a healthy air circulation.
The shape of the AFO is given by an extruded cut. The relative sketch is drawn on the xy plane and is defined by
two splines. The position of its control points is another important parameter. The sketch used for the extruded cut is
shown in Fig. 5. After filleting every sharp edge, the orthosis is ready to be optimized.

Fig. 5: (A) Sketch for the extruded cut and (B) result after applying this feature.
 A. Dal Maso, F. Cosmi / Materials Today: Proceedings 12 (2019) 252–261 257

2.4. Topological optimization

The Ankle-Foot Orthosis will have to resist to certain loads during its life, therefore it must be designed
accordingly. A FEM simulation is carried out on the model, which is then optimized through successive
approximations. The loads to be applied in the FEM simulation depend on the specific function of the device, which
in this case study is merely to sustain the foot during the toe-off and heel-contact phases of each step. To simulate
this, a force of 50 N is applied on the toe-end of the AFO model, while the opposite end is fixed (Fig. 6). In reality,
the distribution of contact forces is far more complex and difficult to determine with precision, however 50 N are
considered sufficiently precautionary in this application.

Fig. 6: Definition of loads and constraints for the FEM simulation.

The AFO is printed in PLA (Poly-Lactic Acid), a common 3D-printing material, biocompatible and non-toxic.
The non-isotropic mechanical characteristics of this 3D-printed material are known [6-9]. The values of maximum
tensile stress and Young’s modulus found in literature were confirmed through tensile tests performed in-house.
The device is printed on a Builder Premium Medium™ 3D printer. Due to the limited size of the buildplate, the
orthosis must be printed upright, i.e. with its z axis parallel to the z axis of the machine. This choice affects the final
geometry, since the mechanical resistance in the z direction is definitely lower than that of the x and y directions in
FDM-printed parts.
The FEM studies are performed in SolidWorks Simulation™. The characteristics of the material defined in
Simulation™ are listed in Table 2. The simulation setup and main results are shown in Fig. 7. These final results
apply to the fully-optimized AFO: to obtain this solution, several intermediate simulations were performed to adjust
the geometrical parameters.

Table 2. Anisotropic material characteristics used for the FEM simulation.

Young’s Modulus in x Ex 3.0 GPa
Young’s Modulus in y Ey 3.0 GPa
Shear Modulus yOz plane τyz 1.1 GPa
Shear Modulus xOz plane τxz 1.1 GPa
Shear Modulus xOy plane τxy 0.7 GPa
Young’s Modulus in z Ez 1.9 GPa
Poisson’s ratio ν 0.36
258 A. Dal Maso, F. Cosmi / Materials Today: Proceedings 12 (2019) 252–261

Fig. 7: Results of the simulations: all 6 components of the stress are plotted.

Due to the transverse isotropy of the material, an appropriate failure criterion must be chosen to decide whether
the part is sufficiently robust. The Tsai-Wo criterion for transversally isotropic materials [10] could be used,
however it requires the knowledge of maximum shear stress for the material, as well as maximum compression
stress, which were not available. Therefore, a different failure criterion was adopted, taking advantage of the fact
that the stress tensor in the critical points is practically planar, since they are located on the AFO surface. It involves
the following steps.

Fig. 8: Stress tensor graphical representation in the critical points.

 At first we determine the critical points, i.e. the points with maximum tensile stress.
 In these points we analyze the stress tensor (Fig. 8). If the direction of the first principal stress ( ) is parallel to
one of the main axes, the local stress is compared directly to the maximum stress of the material in that direction.
 If, instead, the direction of the first principal stress is not parallel to the axes, this direction will be tilted of an
angle γ from the z axis; the stress will then be compared to the maximum stress measured on specimens that were
printed with the same tilt angle γ. These tensile tests are performed in-house with the available tensile testing
machine.
 The point is structurally verified if:
 A. Dal Maso, F. Cosmi / Materials Today: Proceedings 12 (2019) 252–261 259

, ,
= · =0
,
=0

Table 3 summarizes the results for the three main critical points of our case study. From our tests and in literature
we obtained the values for maximum stress measured ( ) in the tested directions. The maximum admissible
stress ( is calculated by applying a safety factor = 2.

Table 3. FEM results in the three critical points.

Point 1 Point 2 Point 3

Tilt angle (referred to the z axis) 30° 45° 90°
Maximum stress measured 20 MPa 25 MPa 45 MPa
Maximum admissible stress 10 MPa 12.5 MPa 22.5 MPa
First principal stress (calculated) 9.2 6.9 8.7

A mesh convergence study is carried out as well, Fig. 9. Other critical points were ignored, since the local high
stress values were clearly due to singularities.

Fig. 9: Mesh convergence in the three critical points.

3. Results.

The designed AFO is finally printed and tested on the patient.

3.1. 3D print

The device is printed on a Builder Premium Medium™ machine. The slicer used is Cura™ v.3.2.1. Table 4
shows the values for the main printing parameters. The whole print took 41 hours and used 143 meters of filament
(corresponding to 430 grams of material), including supports. Following supports removal the and a minimal post-
processing, the AFO weights 250 grams and is ready to be tested on the patient.
260 A. Dal Maso, F. Cosmi / Materials Today: Proceedings 12 (2019) 252–261

Table 4. Parameters set in Cura™ for the 3D print.

Layer Height 0.2 mm Flow 100%
Top/Bottom Line Width 0.4 mm Retraction Distance 6.5 mm
Wall Thickness 1.8 mm Print Speed 50 mm/s
Top/Bottom Thickness 1.8 mm Travel Speed 100 mm/s
Infill Density 100% Support Overhang Angle 20°
Infill Pattern Grid Support Pattern ZigZag
Printing Temperature 210°C Support Density 15%
Preheated Bed Temperature 60°C Brim Width 5 mm

3.2. Test on patient

The only way to evaluate the actual comfort of the device is to have the patient test it (Fig. 10). A great
geometrical correspondence between the foot and the orthosis is immediately noticed and a great comfort is
highlighted. The closing mechanism proves simple and effective, therefore no other additional parts are required.
The patient is overall satisfied by the device.

Fig. 10: AFO tested on the patient.

4. Discussion and conclusions

The presented procedure has many advantages. First of all, versatility: this method can be applied to a wide
variety of cases. Not only can it be adapted to each single patient, but it can be applied also to other parts of the
body. Secondly, time and costs. FDM printing is an extremely cheap production method, that allows to obtain fully-
functional prototypes without large investments. The following formula can be adopted to estimate the 3D printing
production costs[1]:

=
where:
is the total cost of the device;
is the equipment ownership cost, appraised at E=1.50 €/h 41h/print=61.50 €;
is the cost of the material filament, in this case M=50 €/kg x 0.43 kg/print=21.50 €.
is the cost of labor, that can be estimated as L=50.00€/h x 6h/AFO= 300.00 €.
 A. Dal Maso, F. Cosmi / Materials Today: Proceedings 12 (2019) 252–261 261

The total cost of the device can be therefore be estimated at = 383.00 €, which is definitely much lower than
what it would cost if this perfectly customized AFO were to be produced with other technologies.
On the other hand, there are certain limitations, which ought to be improved in future developments.

 Material. PLA is an excellent material for FDM, but it is not the most mechanically resistant. Other materials,
such as Nylon, could allow thinner profiles maintaining the same mechanical characteristics.
 Automation. This method shows some instability when converting the STL file into SolidWorks™ part: this
phase requires some experience and ability, and cannot be easily automated. Another software could be used for
the purpose (for example the freeware Blender™) but that would imply losing the excellent parametrization that
SolidWorks™ allows. Given that the cost of labor is the largest contribution to the total cost of the AFO, this
point must be addressed with special care.
 Optimization. Only a general optimization was carried out: the proposed geometry was not the absolute optimum.
A complete topological optimization could be easily carried out taking advantage of the proposed
parametrization.

Other future developments include the design of an AFO that actively corrects certain orthopaedical defects by
holding the foot in a desired position, for example in pre-pronation. Moreover, designing a 3D-printed mold for the
production of the AFO through injection molding would overcome all the issues related to the anisotropy of 3D-
printed parts, but the cost of the mold production could elevate the AFO cost. Finally, in order to validate
scientifically the efficiency of the 3D-printed AFO, a series of medical tests should be carried out on different
patients and these effects should be quantified and measured.

Acknowledgements

We thank Sara Segantin for having participated as a patient to this case study.

References

[1] I. Gibson, D.W. Rosen, B. Stucker, Additive Manufacturing Technologies - Rapid Prototyping to Direct Digital Manufacturing, Springer,
2010.
[2] Y.H. Cha, K.H. Lee, J.J. Ryu, I.W. Joo, A. Seo A, D.H. Kim, S.J. Kim, Appl. Bionics. Biomec. (2017) doi: 10.1155/2017/9610468
[3] A. Mancini, C. Morlacchi, Clinica Ortopedica, Piccin, 2005, pp. 259-265.
[4] H.L. Mitchell, I. Newton, J Photogramm Remote Sens. 56 (5-6) (2002) 286-294.
[5] Photogrammetry tutorial 11: How to handle a project in Agisoft Photoscan, 2015
<https://dinosaurpalaeo.wordpress.com/2015/10/11/photogrammetry-tutorial-11-how-to-handle-a-project-in-agisoft-photoscan/> Accessed
10/03/2018.
[6] N. Graupner, A.S. Herrmann, J. Müssig, Compos Part A - Appl S. 40 (6-7) (2009) 810-821.
[7] J. S. Bergstrom, D. Hayman, Ann Biomed Eng. 44(2) (2016) 330.
[8] S. Farah. D.G. Anderson, R. Langer, Adv Drug Deliv Rev. 107 (2016) 367-392.
[9] A. Dal Maso, F. Cosmi, Mater Today-Proc. 5(13) (2018) 26739-26746.
[10] S.W. Tsai, E.M. Wu, J. Compos. Mat. 5(1) (1971) 58-80.