This document is downloaded from DR‑NTU (https://dr.ntu.edu.

sg)
Nanyang Technological University, Singapore.

Analysis and modelling of 3D printed springs for

use in spacecraft

Sacco, Enea

2020

Sacco, E. (2020). Analysis and modelling of 3D printed springs for use in spacecraft.
Doctoral thesis, Nanyang Technological University, Singapore.

https://hdl.handle.net/10356/141038

https://doi.org/10.32657/10356/141038

This work is licensed under a Creative Commons Attribution‑NonCommercial 4.0

International License (CC BY‑NC 4.0).

Downloaded on 19 Mar 2024 22:14:58 SGT

ANALYSIS AND MODELLING OF 3D
PRINTED SPRINGS FOR USE IN
SPACECRAFT

ENEA SACCO

A thesis submitted to Nanyang Technological University

in partial fulfilment of the requirement for degree of
Doctor of Philosophy
Supervised by Moon, Seung Ki (Assoc. Prof.)

Mechanical and Aerospace Engineering

Nanyang Technological University
Singapore
June 2, 2020
 Statement of Originality

I hereby certify that the work embodied in this thesis is the result of original

research, is free of plagiarised materials, and has not been submitted for a higher

degree to any other University or Institution.

19th March 2020

................. ...........................
Date Enea Sacco
 Supervisor Declaration Statement

I have reviewed the content and presentation style of this thesis and declare it is

free of plagiarism and of sufficient grammatical clarity to be examined. To the

best of my knowledge, the research and writing are those of the candidate except

as acknowledged in the Author Attribution Statement. I confirm that the

investigations were conducted in accord with the ethics policies and integrity

standards of Nanyang Technological University and that the research data are

presented honestly and without prejudice.

19th March 2020

................. ...........................
Date Asst/P Moon Seung Ki
 Authorship Attribution Statement

This thesis contains material from 1 paper(s) published in the following peer-reviewed
journal(s) / from papers accepted at conferences in which I am listed as an author.

Material from Chapter 2 is published as: Sacco, E., Moon, S.K. Additive manufacturing
for space: status and promises. The International Journal of Advanced Manufacturing
Technology 105, 4123–4146 (2019). https://doi.org/10.1007/s00170-019-03786-z

The contributions of the co-authors are as follows:

 Prof Moon provided the initial project direction and revised the manuscript
drafts.
 I prepared the manuscript drafts.

19th March 2020

................. ...........................
Date Enea Sacco
Acknowledgements
I want to thank my Professor for guiding and advising me throughout this work,
his contribution was invaluable to my research. I also want to thank my girl-
friend and friends for putting up with me for all these years, this work would
have not been possible without you. Special thanks Yunus, Rahul, and Burak for
all your help. Simone, I could not have done it without you. Finally I want to
thank my mother and family for all their love and support.
 CONTENTS

List of Figures

List of Tables

Acronyms

1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Scope and objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Literature review: 3D printing for the space sector 6

2.1 3D printed springs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Research in AM for space . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.1 The importance of mass reductions for spacecraft and launch-
ers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.2 Usage of new technologies in the space sector . . . . . . . . 12
2.2.3 In-space AM . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.4 Ground based AMFS . . . . . . . . . . . . . . . . . . . . . . . 20
2.3 Research in AMFS and 3D printing of springs: gaps and directions 37

3 Proposed research method 41

3.1 Novel CGT design for small spacecraft . . . . . . . . . . . . . . . . . 42
3.1.1 Cold gas thrusters . . . . . . . . . . . . . . . . . . . . . . . . 42
3.1.2 Using AM to increase the dynamic range of CGTs . . . . . . 42
3.1.3 The problem with linear springs . . . . . . . . . . . . . . . . 44
3.2 Deployable helical antennae . . . . . . . . . . . . . . . . . . . . . . . 46
3.3 Research method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.3.1 Material selection . . . . . . . . . . . . . . . . . . . . . . . . . 48
 3.3.2 Comparing springs made via Additive Manufacturing (AM)
and conventional means . . . . . . . . . . . . . . . . . . . . . 50
3.3.3 In-fill selection . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.3.4 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

4 Creation of an orthotropic model for PLA 56

4.1 Determination of material model . . . . . . . . . . . . . . . . . . . . 57
4.1.1 Reference axis and coordinate origin . . . . . . . . . . . . . . 57
4.1.2 Road angle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.1.3 Anisotropic materials . . . . . . . . . . . . . . . . . . . . . . 58
4.2 Material properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.2.1 Young’s modulus E . . . . . . . . . . . . . . . . . . . . . . . . 65
4.2.2 Poisson’s ratio ν . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.2.3 Shear modulus G . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.3 Methodology for measuring material properties . . . . . . . . . . . 69
4.3.1 Sample orientation and naming convention . . . . . . . . . . 69
4.3.2 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.3.3 GOM Correlate . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.3.4 Print settings . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.4.1 Reasons for switching from type IV to type I coupons . . . . 79
4.4.2 Measured constants . . . . . . . . . . . . . . . . . . . . . . . 81

5 Multi-material finite element simulation of 3D printed materials 87

5.1 Modelling of 3D printed materials . . . . . . . . . . . . . . . . . . . 87
5.2 Experimental procedure . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.3 Simulation and experimental verification . . . . . . . . . . . . . . . 89
5.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

6 Exploration of 3D printed springs 94

6.1 Springs and AM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
6.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.2.1 Spring types . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.2.2 Spring wire cross-section . . . . . . . . . . . . . . . . . . . . 95
6.2.3 Printing orientation and in-fill direction . . . . . . . . . . . . 98
6.2.4 Spring index and naming convention . . . . . . . . . . . . . 99
6.2.5 Printer settings . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.2.6 Testing equipment and data generated . . . . . . . . . . . . 101
6.2.7 Tension spring testing . . . . . . . . . . . . . . . . . . . . . . 102
6.2.8 Compression spring testing . . . . . . . . . . . . . . . . . . . 105
6.2.9 Analysis of test data . . . . . . . . . . . . . . . . . . . . . . . 107
6.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.3.1 Tension springs . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.3.2 Compression springs . . . . . . . . . . . . . . . . . . . . . . . 122
7 Derivation of an equation for calculating the spring constant of 3D printed
springs 131
7.1 Existing equation for springs . . . . . . . . . . . . . . . . . . . . . . 132
7.1.1 Calculating the stiffness of isotropic springs . . . . . . . . . 133
7.2 Derivation method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
7.2.1 Warping function vs stress function approach . . . . . . . . 138
7.3 Derivation using warping functions . . . . . . . . . . . . . . . . . . 139
7.3.1 Final solution for the warping function without Fourier series144
7.3.2 Verification of the solution . . . . . . . . . . . . . . . . . . . . 144
7.4 Using Prandtl’s stress function . . . . . . . . . . . . . . . . . . . . . 147
7.4.1 Verification of the solution . . . . . . . . . . . . . . . . . . . . 148
7.4.2 Final derivation of the equation for k . . . . . . . . . . . . . . 151
7.4.3 Verification of the final equation for the stiffness of a 3D
printed spring . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
7.4.4 ULTEM springs . . . . . . . . . . . . . . . . . . . . . . . . . . 157

8 Costs, manufacturing time, material usage, and design freedom 160

8.1 Material usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
8.2 Cost and manufacturing time . . . . . . . . . . . . . . . . . . . . . . 162
8.3 Leveraging the design freedom of 3D printing to reduce the cost
per spring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
8.3.1 Deactivation of support generation . . . . . . . . . . . . . . . 165
8.3.2 Changing the spring’s orientation . . . . . . . . . . . . . . . 165
8.3.3 Increasing the pitch . . . . . . . . . . . . . . . . . . . . . . . . 165
8.3.4 Hollow wire springs . . . . . . . . . . . . . . . . . . . . . . . 166

9 Conclusion 171
9.1 Summary and Conclusions . . . . . . . . . . . . . . . . . . . . . . . 171
9.2 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
9.2.1 Polylactic Acid (PLA) material characteristics . . . . . . . . 174
9.2.2 Multi-material Finite Element Analysis (FEA) of 3D printed
PLA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
9.2.3 3D printing of helical springs . . . . . . . . . . . . . . . . . . 175
9.2.4 Equation for calculating the spring constants of springs man-
ufactured via AM . . . . . . . . . . . . . . . . . . . . . . . . . 176
9.2.5 3D printed springs and design freedom . . . . . . . . . . . . 177
9.3 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.3.1 Multi-material FEA . . . . . . . . . . . . . . . . . . . . . . . . 177
9.3.2 Investigation of 3D printed springs . . . . . . . . . . . . . . 177
9.3.3 Modelling the spring constant of springs manufactured via
AM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
9.3.4 Design freedom . . . . . . . . . . . . . . . . . . . . . . . . . . 178

Appendices 179
A Investigation of localised strain oscillations during slow tensile loading180
A.1 Observation of strain oscillations . . . . . . . . . . . . . . . . . . . . 181
A.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
A.2.1 Tensile coupon specimens . . . . . . . . . . . . . . . . . . . . 186
A.2.2 Analysis of experimental data . . . . . . . . . . . . . . . . . . 188
A.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
A.3.1 Analysis of type I coupons . . . . . . . . . . . . . . . . . . . 193
A.3.2 Analysis of modifed type I coupons in the E3 orientation . . 206

B Derivation of the shear modulus from the torsion constant 209

C Comparison of calculated and experimental spring constants 212

 LIST OF FIGURES

2.1 Springs with four different pitches, four samples were manufac-
tured per pitch value [1]. . . . . . . . . . . . . . . . . . . . . . . . . . 7
a The Ondulé spring design tool interface [2]. . . . . . . . . . 9
b Results showing that 3D-printed helical springs have sim-
ilar twisting performance to theoretical predictions with
varied d, D, N, and L values [2]. . . . . . . . . . . . . . . . . 9
2.3 Example of a leaf spring created using 4D printing. When first
produced the spring was flat but it became curved when the right
conditions were met [3]. . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4 Final velocities as a function of launch vehicle mass ratios calcu-
lated using equation Equation 2.4. . . . . . . . . . . . . . . . . . . . 12
2.5 Some of the objects printed on the International Space Station (ISS)
[4]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.6 The Additive Manufacturing Facility (AMF) mounted in an EX-
PRESS Rack Mid-Deck Locker in the ISS (Image credit: National
Aeronautics and Space Administration (NASA)). . . . . . . . . . . . 15
2.7 The Multipurpose Precision Maintenance Tool (MPMT), created
by a university student as part of the “Future Engineers Space Tool
Challenge”, printed on the ISS [5]. . . . . . . . . . . . . . . . . . . . 17
2.8 Springs printed on the ISS [6]. . . . . . . . . . . . . . . . . . . . . . . 17
2.9 A concept art of SpiderFab Bot creating a truss in orbit [7]. . . . . . 19
2.10 System architecture for in-space 3D Printing (3DP) spacecraft [8]. . 20
2.11 Examples of different configurations of CubeSats [9]. Left - 1U
“Phonesat”; top right - 1.5U “EDSN Spacecraft”; bottom right - 6U
“EcAMSat”. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.12 Comparison of Selective Laser Melting (SLM) build volumes to
rocket engines [10]. Dimensions in SI starting from the left are
25.4 × 25.4 × 25.4 cm (16387.064 cm3 ), 39.37 × 60.96 × 48.26 cm
(115823.768 cm3 ), 228.6 cm, 116.84 cm, 177.8 cm, and 142.24 cm. . . 25
2.13 Tim Ellis, co-founder of Relativity, holding a printed nozzle in
front of the Stargate printing system [11]. . . . . . . . . . . . . . . . 25
2.14 Schematic of a Nuclear Thermal Propulsion (NTP) rocket [12]. Fuel
is pumped into the combustion chamber where a nuclear reactor
provides the heat for combustion. This is much more efficient than
traditional rockets due to the far greater amount of thermal energy
provided by the nuclear reactor. . . . . . . . . . . . . . . . . . . . . . 26
2.15 Titanium brackets manufactured using an EOSINT M 280 for use
on Eurostar E3000 satellites [13]. . . . . . . . . . . . . . . . . . . . . 28
2.16 The Adel’Light hinge for deployment of solar panels [14]. . . . . . . 29
2.17 Artist’s impression of the thruster in a 3U CubeSat [15]. . . . . . . . 31
2.18 Diagram showing the tanks, pipes, nozzles, and valves for the
thruster [16]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.19 Overview of Automated Additive Construction System (ACES)-3
[17]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.20 Solar powered 3D printer with xenon lamps for testing [18]. . . . . 34
2.21 Outpost structure (top) and wall profile (bottom left) with detail
(bottom right) [19]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3.1 Examples of types of springs [20]. . . . . . . . . . . . . . . . . . . . . 42

3.2 Example Cold Gas Thruster (CGT) module design schematic show-
ing the main components: propellant tank, plena, valves, and noz-
zle [21]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3 A hypothetical CGT that uses 3D printed springs. . . . . . . . . . . 43
3.4 The spring at rest position is defined as x = 0. When a force is
applied the displacement from the initial position is called x. . . . . 44
3.5 Example of a non-linear, dual pitch spring [22]. . . . . . . . . . . . . 45
3.6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
a Packaging schemes for CubeSat antennae: (a) the helical
pantograph, (b) coilable conductors, (c) dual-matrix com-
posite shells, (d) hinged ribs, and (e) the wrapped mesh [23]. 47
b Simulation of compaction of eight-helix pantograph antenna
[24]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.7 Cross-section of a part printed with Fused Filament Fabrication
(FFF) showing various features. . . . . . . . . . . . . . . . . . . . . . 51
3.8 Comparison of ±45 and mono-directional in-fills. . . . . . . . . . . 53
a Cross-section ±45 in-fill changes along the coil therefore
the shear constants do so too. . . . . . . . . . . . . . . . . . . 53
b Mono-directional infill has a constant cross-section through-
out the coil. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.9 Thesis roadmap. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

4.1 Definition of printer axes and coordinate origin [25]. . . . . . . . . . 57

4.2 Cross-section of a part printed with FFF showing various features
including road angle [26]. . . . . . . . . . . . . . . . . . . . . . . . . 58
4.3 Definition of shear strain. . . . . . . . . . . . . . . . . . . . . . . . . 59
4.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
 a Scanning Electron Microscope (SEM) image of material ex-
trusion printed plastic that shows the gaps between tracks
[27]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
b Cross-section of material extrusion printed Acrylonitrile Bu-
tadiene Styrene (ABS) that shows the air gaps produced
during the printing process [28]. . . . . . . . . . . . . . . . . 60
4.5 Typical stress-strain curve for a material. From A to B the material
obeys Hooke’s law (elastic behaviour). At C it starts to yield, from
C to D it undergoes necking, and finally at E the material breaks. . 61
4.6 Microstructural details of a 3D printed material. . . . . . . . . . . . 62
4.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
a Labelling of normal stresses, shear stresses, and material
directions from ASTM standard 5379 [29]. . . . . . . . . . . . 64
b Definition of material directions used in this research, where
1 and 2 are aligned to the tracks and 3 is parallel to the de-
position direction. . . . . . . . . . . . . . . . . . . . . . . . . 64
4.8 ASTM tensile coupons used [30]. . . . . . . . . . . . . . . . . . . . . 66
4.9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
a Sample used for shear testing [29]. . . . . . . . . . . . . . . . 68
b Testing fixtures used to apply shear stress [29]. . . . . . . . . 68
4.10 Shear stress against shear strain showing important features from
ASTM standard D5379 [29]. . . . . . . . . . . . . . . . . . . . . . . . 69
4.11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
a Orientation of material planes used to develop coupon ori-
entations [29]. . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
b Diagram of sample orientations naming system. The white
axes shows the printer coordinate system while the blue
axes show the material directions. Although type IV sam-
ples are shown here, type I were also printed in the same
orientations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.12 Type I tensile coupon ready with black lines for measuring axial
strain. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
a Example of selection of the area for analysis with pattern
quality legend on the right. . . . . . . . . . . . . . . . . . . . 75
b Simplified view of the analysis of sample ±45-E3-07 in GOM
Correlate indicating significant features. . . . . . . . . . . . . 75
c Strain analysis for sample +45-G13-02. The measurement
area can be seen in between the two notches. The x and y
axes have been rotated to produce the required shear strain
as described in ASTM standard D5379. . . . . . . . . . . . . 75
4.14 Output from the strain analysis in GOM Correlate of a tensile coupon
showing the average axial (blue) and transverse (black) strains
over time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.15 Plots of data used to calculate E1 . . . . . . . . . . . . . . . . . . . . . 77
a Complete stress-strain plots for all samples. . . . . . . . . . 77
 b Plots of linear portions of stress-strain curves showing av-
erage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
a Stress-strain curves produced by tensile testing of 3D printed
PLA at different raster orientations [31]. . . . . . . . . . . . . 78
b Raster orientation directions, 0o (top), 90o (middle), and 45o
(bottom) [31]. . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.17 Original Prusa i3 MK3 [32]. . . . . . . . . . . . . . . . . . . . . . . . 78
4.18 Plot for measuring Poisson’s ratio using sample ±45-E3-03, a type
IV coupon. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
a Time versus axial strain. . . . . . . . . . . . . . . . . . . . . . 80
b Time versus transverse strain. . . . . . . . . . . . . . . . . . . 80
c Transverse versus axial strain. . . . . . . . . . . . . . . . . . 80
4.19 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
a Transverse versus axial strain for all ±45-E3 samples using
type IV coupons. . . . . . . . . . . . . . . . . . . . . . . . . . 84
b Transverse versus axial strain for all ±45-E3 samples using
type I coupons. . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.20 Comparison of measured mechanical properties for ±45 and +45
in-fill using PLA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
a Elastic modulus. . . . . . . . . . . . . . . . . . . . . . . . . . 85
b Poisson’s ratio. . . . . . . . . . . . . . . . . . . . . . . . . . . 85
c Shear modulus. . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.21 Comparison of contact regions between layers in +45 and ±45 in-
fills. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

5.1 Cross-section of sliced object with various parts coloured in. The
grey cylinder on the right represents the nozzle head. . . . . . . . . 88
5.2 Abaqus assembly showing the in-fill and perimeters. . . . . . . . . 89
5.3 Top down of sample to show material directions for the perimeter
(red) and the in-fill (blue). Direction 3 for both is out of the page. . 90
5.4 Comparison of simulated and printed samples. . . . . . . . . . . . . 92
a Elastic modulus. . . . . . . . . . . . . . . . . . . . . . . . . . 92
b Poisson’s ratio. . . . . . . . . . . . . . . . . . . . . . . . . . . 92

6.1 Spring parameters, Na refers to the active coils of the spring, those
are coils that are not part of either end of the spring. . . . . . . . . . 96
6.2 Comparing the print quality of circular and square wire cross-
sections. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
a Circular cross-section. . . . . . . . . . . . . . . . . . . . . . . 97
b Square cross-section. . . . . . . . . . . . . . . . . . . . . . . . 97
6.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
a Spring with printed grips. . . . . . . . . . . . . . . . . . . . . 98
b Spring with screws as grips in the testing machine. . . . . . 98
6.4 Material orientations of a single coil for FEA. . . . . . . . . . . . . . 99
 a ±45 in-fill. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
b Mono-directional in-fill. . . . . . . . . . . . . . . . . . . . . . 99
6.5 Stress distributions generated by a constant upwards velocity of
0.5 mm/min. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
a Mono-directional in-fill. Uneven meshing produced areas
of higher stress which can be ignored. . . . . . . . . . . . . . 99
b ±45 in-fill. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.6 Example of a PLA compression spring. . . . . . . . . . . . . . . . . . 100
6.7 Compression springs in the slicing program. . . . . . . . . . . . . . 102
a Mono-directional in-fill. . . . . . . . . . . . . . . . . . . . . . 102
b ±45 in-fill. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.8 Monotonic and cyclic stress-strain curves for SAE 5160 spring steel
[33]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.9 F-x plot for all ±45 tension springs. Sets of four springs with the
same dimensions have the same colour. . . . . . . . . . . . . . . . . 108
6.10 Comparison between torsion tests and FEA simulations in various
test pieces [34]. The plots of the torsion tests are grey in colour. . . . 110
a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
c . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.11 Noise present in the testing. . . . . . . . . . . . . . . . . . . . . . . . 113
6.12 Results of the cyclical testing in the pseudo-linear region. . . . . . . 114
6.13 Force-displacement curves of d5D25 mono spring to breakage. . . . 115
6.14 Screenshot of the Curve Fitting Toolbox (CFT) in MATLAB to show
where k2 was measured. The red portions of the plot are ignored
by the straight line fitting. . . . . . . . . . . . . . . . . . . . . . . . . 116
6.15 Cyclical testing of d5D25 tension mono spring. Small displace-
ment: 0.12 - 0.08 m, large displacement: 0.095 - 0.105 m. . . . . . . . 117
6.16 F-x plot of all d5D25 springs tested to fracture. . . . . . . . . . . . . 118
6.17 Example of ±45 in-fill resulting in a greater number of voids. . . . 119
6.18 Breakage comparison between a ±45 spring (left) and a mono spring
(right). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.19 Fractured ends of springs. . . . . . . . . . . . . . . . . . . . . . . . . 121
a ±45 spring, the indicated lighter areas are fractures along
the layers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
b Mono spring, where the fracture plane is perpendicular to
the coil. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.20 One of the d5D20P15 compression mono springs tested to com-
pare to the tension springs with the same dimensions. . . . . . . . . 122
6.21 Viscoelastic creep during cyclical testing of two weak mono PLA
compression springs. . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.22 Plot of cyclical testing of ULTEM and PLA springs with creep re-
moved. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
6.23 Viscoelastic creep in cyclical testing. . . . . . . . . . . . . . . . . . . 128
a Comparison of ULTEM 9085 and PLA springs. . . . . . . . . 128
b ULTEM 9085 springs. . . . . . . . . . . . . . . . . . . . . . . 128
7.1 Change in wire cross-section after coiling for rectangular wire springs
[35]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
7.2 Example of simplification of parts using machined springs [36]. . . 133
a Three piece part with wire wound spring. . . . . . . . . . . . 133
b Single part with machined spring. . . . . . . . . . . . . . . . 133
7.3 Distance that coils move when a spring is stretched. . . . . . . . . . 134
a Side view. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
b Front view. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
7.4 Rectangular prismatic bar under torsion at one end. . . . . . . . . . 138
7.5 Shear moduli for ideal mono in-fill where 1 is the direction tangen-
tial to the coils, 3 is parallel to layer deposition, and 2 is orthogonal
to the other two. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
7.6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
a Plot of the warping function φ( x, y) without using Fourier
series. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
b Plot of the derivatives of the warping function where or-
ange is φx ( x, y) and blue is φy ( x, y). . . . . . . . . . . . . . . 145
7.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
a Values of GJn for each n. . . . . . . . . . . . . . . . . . . . . . 146
b Sum of GJn as n increases. . . . . . . . . . . . . . . . . . . . . 146
7.8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
a Plot of the Prandtl stress function. . . . . . . . . . . . . . . . 149
b Plot of the shear stresses. . . . . . . . . . . . . . . . . . . . . 149
7.9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
a Values of GJn for each n. . . . . . . . . . . . . . . . . . . . . . 150
b Convergence of GJn as n increases. . . . . . . . . . . . . . . . 150
7.10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
a Ideal wire cross-section. . . . . . . . . . . . . . . . . . . . . . 152
b Wire cross-section taken from (c). . . . . . . . . . . . . . . . . 152
c Side view of a more realistic uncoiled spring. . . . . . . . . . 152
7.11 Printing of springs using robotic arms. . . . . . . . . . . . . . . . . . 153
7.12 Plots of the measured and calculated spring constants. . . . . . . . 155
a With the shear realignment. . . . . . . . . . . . . . . . . . . . 155
b Without the shear realignment. . . . . . . . . . . . . . . . . . 155
a Change in shear modulus with infill density [2]. . . . . . . . 158
b Experimental (blue) vs calculated (orange) spring stiffness
[2]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
c . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

8.1 Normalised cost breakdown for springs from Table 8.2. . . . . . . . 164
8.2 Springs in Simplify3D showing that no orientation is possible that
does not require supports. The printer axes are shown for reference.166
8.3 Support generation as θrise is increased. . . . . . . . . . . . . . . . . 167
a θrise = 45o . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
b θrise = 63o . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
c θrise = 76o . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
8.4 Second moment of area for a hollow tube with outer radius 2 mm. . 168
8.5 Hollow springs with a wall thickness of 2 mm. . . . . . . . . . . . . 168
a The three cross-sections considered: triangle (top), rhom-
bus (middle), square (bottom). . . . . . . . . . . . . . . . . . 168
b Equilateral triangle wire cross-section. ULTEM 9085 needed:
60.93 cm3 ; support: 58.27 cm3 ; support/material ratio: 0.95. 168
c Rhombus wire cross-section. ULTEM 9085 needed: 55.23
cm3 ; support: 31.01 cm3 ; support/material ratio: 0.56. . . . . 168
8.6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
a Change in GJn as thickness of hollow square tube decreases. 169
b Compression spring with hollow, square wire printed out
of PLA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

A.1 Type IV sample in the Digital Image Correlation (DIC) program.

The red area corresponds to the gauge area and is where the strain
was measured. The green arrow (y-axis) shows the direction of
axial strain while the red one (x-axis) shows the direction of trans-
verse strain. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
A.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
a Example of a deviation label, identified by the green label,
on the surface of sample ±45-E3-07. . . . . . . . . . . . . . . 184
b Average strain in the y direction compared to the strain for
the point shown in (a), also in the y-direction, for sample
±45-E3-07. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
A.3 Examples of point oscillations from tensile samples in different
orientations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
a E1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
b E2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
c E3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
A.4 Strains for a single point on sample ±45-E3-07, where the axes are
aligned like in Figure 4.13b. . . . . . . . . . . . . . . . . . . . . . . . 185
a x (transverse) direction. . . . . . . . . . . . . . . . . . . . . . 185
b y (axial) direction. . . . . . . . . . . . . . . . . . . . . . . . . . 185
A.5 Modified type I coupon details. . . . . . . . . . . . . . . . . . . . . . 187
a Example of black markings on the right of the sample to
emphasize the layers. . . . . . . . . . . . . . . . . . . . . . . 187
b Testing on the Universal Testing Machine (UTM). . . . . . . 187
A.6 Example of a series of deviation labels aligned to layers (left) and
their strain-time plots (right), where they are all shown together. . . 188
A.7 Example of a series of 5 deviation labels aligned to layers. . . . . . . 190
A.8 Plots of displacement and strain for 5 deviation labels on the sur-
face of sample ±45-E3-01. . . . . . . . . . . . . . . . . . . . . . . . . 191
a Displacement over time in the axial direction. . . . . . . . . 191
b True strain over time in the axial direction. . . . . . . . . . . 191
A.9 Example of processing of noisy strain-time data within MATLAB. . 192
A.10 Example of peaks found on a filtered and de-linearised signal, pri-
mary (green) and secondary peaks (highlighted in red) are shown. 193
A.11 Fitting a sinusoidal function using the CFT. . . . . . . . . . . . . . . 194
A.12 Strain-time plots for strain oscillations observed in 10 adjacent lay-
ers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
a Layer 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
b Layer 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
c Layer 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
d Layer 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
e Layer 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
f Layer 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
g Layer 7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
h Layer 8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
i Layer 9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
j Layer 10. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
A.13 Example of fitting Equation A.1 to the strain-time data of a layer. . 197
A.14 Example of two adjacent layers that have different oscillation am-
plitudes. The plot highlighted in red corresponds to the strain pro-
duced by the deviation label also highlighted in red, while the plot
below the highlighted one corresponds to the deviation label just
below the highlighted one. . . . . . . . . . . . . . . . . . . . . . . . . 200
A.15 Analogue amplitude modulation. . . . . . . . . . . . . . . . . . . . . 201
A.16 Strain oscillations seen in GOM from deviation labels placed in
regions of alternating striations. . . . . . . . . . . . . . . . . . . . . . 202
a Deviation points in adjacent striation regions. . . . . . . . . 202
b Point 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
c Point 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
d Point 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
A.17 Example of striations observed in this research compared to Bar-
tolai et al. The strain fields in (a) and (b) are not superimposed on
the video of the samples. . . . . . . . . . . . . . . . . . . . . . . . . . 204
a ±45-E3-01. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
b ±45-E1-01. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
c Full field strain images of strain in the loading direction
for 100% infill density with ±45 infill with ±45 deg (left),
+30 deg / − 60 deg (middle), and 0 deg /90 deg (right) tool-
path orientations [37]. . . . . . . . . . . . . . . . . . . . . . . 204
A.18 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
a Dimensions of large tensile specimen with a thickness of
19 mm [38]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
b DIC strain fields taken the instant before sample rupture [38].205
A.19 Strain oscillations produced by modified type I coupons. . . . . . . 207
a . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
c . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
A.20 Fitting Equation A.2 on the signal produced by modified type I
coupons in MATLAB’s CFT. . . . . . . . . . . . . . . . . . . . . . . . 208

B.1 Cylinder before torsion (left) and after torsion (right). . . . . . . . . 210
 LIST OF TABLES

2.1 Cost to launch to Low Earth Orbit (LEO) using the Space Launch
System and Falcon Heavy (FH) [39], all costs are in USD. . . . . . . 13
2.2 AMF system specifications [40]. . . . . . . . . . . . . . . . . . . . . . 15
2.3 Results from tests on samples prepared in orbit (0 g) and on the
ground (1 g) [41, 42]. . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Developing a prototype rocket engine with AM versus conven-
tional manufacturing [43]. . . . . . . . . . . . . . . . . . . . . . . . . 23
2.5 Constituents of the samples investigated in the Additive Construc-
tion with Mobile Emplacement (ACME) study [44]. . . . . . . . . . 34

4.1 Dimensions of type I and IV coupons [30]. . . . . . . . . . . . . . . . 65

4.2 Dimensions of shear sample [29]. . . . . . . . . . . . . . . . . . . . . 67
4.3 Print settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.4 Measured orthotropic constants for +45 and ±45 in-fill using type
I and shear coupons made with PLA. . . . . . . . . . . . . . . . . . . 81

5.1 Elastic moduli and Poisson’s ratios from simulations and experi-
ments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.2 Percentage differences between the experimental and simulated
elastic moduli and Poisson ratios . . . . . . . . . . . . . . . . . . . . 93

6.1 Dimensions of PLA tension springs. . . . . . . . . . . . . . . . . . . 103

6.2 Dimensions of PLA ±45 compression springs. . . . . . . . . . . . . 105
6.3 Dimensions of PLA mono compression spring. . . . . . . . . . . . . 106
6.4 Dimensions of compression ULTEM 9085 springs . . . . . . . . . . 107
6.5 Measured spring constants, average and standard deviations for
±45 tension springs stretched to breakage. All springs have P = 7
and so it has been omitted. . . . . . . . . . . . . . . . . . . . . . . . . 109
6.6 ±45 tension springs ranked according to their measured k1 . . . . . 111
6.7 Results of load rate verification tests for ±45 tension spring d4D45 112
6.8 Results of load rate verification tests for ±45 tension spring d5D45 113
6.9 Measured k1 s and k2 s for d5D25 mono spring to breakage. . . . . . 116
6.10 Comparison of tension and compression d5D20P15 springs. . . . . 122
6.11 ks for compression springs with ±45 and mono in-fills. . . . . . . . 123
6.12 Measured stiffness for mono compression springs. . . . . . . . . . . 124
6.13 Spring constants measured from cyclical testing of 5 d5D20P15
and 3 d5D20P15U springs. . . . . . . . . . . . . . . . . . . . . . . . . 126
6.14 Comparison of ULTEM 9085 and PLA springs. . . . . . . . . . . . . 127
6.15 Effects of layer height variation on the stiffness of PLA springs,
first round of testing with one d5D20P15 per layer height. . . . . . 129
6.16 Effects of layer height variation on the stiffness of PLA springs;
second round of testing with three d7D20P15 springs per layer
height. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

7.1 Spring constants for ±45 in-fill with shear correction. . . . . . . . . 156
7.2 Spring constants ±45 in-fill without shear correction. . . . . . . . . 157
7.3 Stiffness calculated using Equation 7.46 with no shear correction
and G12 , G13 = 800 MPa. . . . . . . . . . . . . . . . . . . . . . . . . . 157
7.4 Measured shear constants for ULTEM 9085 compared to PLA. . . . 158
7.5 Comparison of calculated and experimental stiffness of ULTEM
springs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

8.1 Ratio of support material to ULTEM 9085 used in printing cylin-

drical springs and resultant cost. The cost of both ULTEM 9085
and support used for calculations is 0.87 SGD/cm3 . . . . . . . . . . 161
8.2 Estimated cost of printing springs with ULTEM 9085 using the
method from Atzeni et al. [45]. . . . . . . . . . . . . . . . . . . . . . 163
8.3 Comparison of costs for square hollow springs with different shell
thicknesses t. The machine cost per hour is 56.90 SGD and the cost
of the ULTEM and support is 0.87 SGD/cm3 . . . . . . . . . . . . . . 169

A.1 Dimensions of redesigned type I coupon compared to the regular

type I (the dimensions referenced can be found in Figure 4.8). . . . 186
A.2 Mean amplitudes and periods of strain oscillations in each layer
of a Type I coupon. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
A.3 Curve fitting for 10 layers using a single sine term. . . . . . . . . . . 197
A.4 Curve fitting for 10 layers using two sine terms. . . . . . . . . . . . 199
A.5 Quality of curve fitting for modified type I coupons. . . . . . . . . . 208

C.1 Calculated and measured spring constants with the shear modu-
lus realignment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
C.2 Calculated and measured spring constants without the shear mod-
ulus realignment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
 ACRONYMS

3DP 3D Printing 10, 13, 18, 20, 24, 28, 29, 36, 37, 72
3DPrint 3D Printing in Zero-G Experiment 14
ABS Acrylonitrile Butadiene Styrene 14–16, 18, 29, 30, 49, 50, 60, 82, 127, 162, 201,
203
ACES Automated Additive Construction System 33
ACME Additive Construction with Mobile Emplacement 33, 34
ALD Atomic Layer Deposition 26
AM Additive Manufacturing 1–3, 5, 6, 8–10, 12–14, 16, 18, 20, 22–24, 26–31, 34,
36–41, 45, 50, 53, 58, 60, 65, 67, 94, 95, 111, 132, 137, 160–162, 164, 171,
172, 175, 176, 178
AMAZE Additive Manufacturing Aiming Towards Zero Waste and Efficient Pro-
duction of High-Tech Metal Parts 26, 27
AMF Additive Manufacturing Facility 14–16
AMFS Additive Manufacturing For Space 2, 5, 6, 10, 14, 20, 22, 24, 26–28, 32, 38,
39, 41, 172
AMSC America Makes & ANSI Additive Manufacturing Standardization Col-
laborative 38
AMSII Additive Manufacturing Structural Integrity Initiative 39
BC Boundary Condition 140–143, 146–148
BJP Binder Jet Printing 36
BVP Boundary Value Problem 139
CAD Computer Aided Design 97
CFT Curve Fitting Toolbox 107, 116, 193, 194, 196, 208
CGT Cold Gas Thruster 3, 42, 43, 172
CIRAS Commercial Infrastructure for Robotic Assembly and Services 19
CT Computer Tomography 14, 23
DED Direct Energy Deposition 24, 27
DFAM Design for Additive Manufacturing 37, 38, 172
DIC Digital Image Correlation 72–74, 83, 173, 182, 183, 186, 203, 205
DLP Digital Light Processing 1
DMLS Direct Metal Laser Sintering 26, 28, 29
DMP Direct Metal Printing 36
EBF3 Electron-Beam Freeform Fabrication 24
EBM Electron Beam Melting 9, 23, 24
EOS Elctro Optical Systems 48
ESA European Space Agency 10, 35
FDM Fused Deposition Modeling 49, 57
FEA Finite Element Analysis 52, 58, 88–90, 93, 98, 99, 110, 137, 175, 177
FFF Fused Filament Fabrication 3, 14, 16, 18, 19, 29, 30, 34, 36, 49, 51, 58, 82, 88,
108, 111, 136, 154, 173
FH Falcon Heavy 13
GRC Glenn Research Center 22, 23, 26
GSFC Goddard Space Flight Center 26
HCF High-Cycle Fatigue 23
HIP Hot Isostatic Pressing 22, 23
ISAM In-Space Additive Manufacturing 14, 18, 172
ISM In-Space Manufacturing 14, 17–19, 172
ISRU In-Situ Resource Utilization 32–36, 172
ISS International Space Station 14–18, 22, 29, 172
JPL Jet Propulsion Laboratory 32
LBM Laser Beam Melting 23
LCUSP Low Cost Upper Stage-Class Propulsion 24
LENS Laser Engineered Net Shaping 2, 24, 35, 153
LEO Low Earth Orbit 13, 24, 49
LFMT Laser Freeform Manufacturing Technology 24
LOM Laminated Object Manufacturing 36
MIS Made In Space 14, 16, 19
MOV Main Oxidizer Valve 28
MPMT Multipurpose Precision Maintenance Tool 16, 17
MSFC Marshall Space Flight Center 14, 26
MSFC-STD-3716 Engineering and Quality Standard for Additively Manufac-
tured Spaceflight Hardware 39
MSFC-STD-3717 Specification for Control and Qualification of Laser Powder
Bed Fusion Metallurgical Processes 39
MSG Microgravity Science Glovebox 14
NASA National Aeronautics and Space Administration 10, 13–15, 18, 22, 24, 28,
30, 32, 33, 39, 43
NRC National Research Council 12
NTP Nuclear Thermal Propulsion 26
ODE Ordinary Differential Equation 148
OSIRIS-REx Origins, Spectral Interpretation, Resource Identification, Security,
Regolith Explorer 12, 13
PBF Powder Bed Fusion 22, 27, 36, 39
PC Polycarbonate 14, 15, 30, 49
PDE Partial Differential Equation 139, 140, 147
PE Polyethylene 14, 15
PEEK Polyetheretherketone 32, 50
PEI Polyetherimide 50
PETG Polyethylene Terephthalate Glycol 49
PLA Polylactic Acid 3, 5, 7, 30, 49, 50, 52–54, 58, 74, 76, 78, 81–83, 85, 95, 100, 103,
105–108, 112, 114, 117, 125–130, 154, 157–159, 161, 162, 169, 173–177, 201,
205
PPT Pulsed Plasma Thruster 30
RFP Rapid Freeze Prototyping 1
SEM Scanning Electron Microscope 60
SLA Stereolithography 1, 31
SLM Selective Laser Melting 1, 7, 22, 24, 25, 35, 37, 39, 48
SLS Selective Laser Sintering 1, 32, 34, 167
TAS Thales Alenia Space 10, 29
TPU Thermoplastic Polyurethane 49, 50
TUI Tethers Unlimited, Inc. 18
UTM Universal Testing Machine 71, 72, 76, 81, 90, 97, 101, 102, 107, 112, 187, 188
Summary
The use of additive manufacturing in the space industry is growing and there are
many developments being made in a wide variety of fields, from tools for astro-
nauts to rocket combustion chambers to habitats on other worlds. Yet there are
still several knowledge gaps that slow the development of standards and there-
fore the widespread use of 3D printing in the space industry. One of the gaps is
the application of 3d printing to springs for use in spacecraft mechanisms. Con-
sequently, the objective of this research is to analyse and model the behaviour
of these widespread components. The first step is to characterise the material,
which in this case is PLA printed using material extrusion with two different
in-fills, using the orthotropic model. Another material, ULTEM 9085, was also
used later in the research. The results showed that the two in-fill are mechani-
cally similar but one is slightly better under shear stresses. Based on the under-
standing of the material characteristics, simulations are conducted using tensile
coupons with varying numbers of perimeters and then compared to samples
printed with the same characteristics. The simulations and experimental results
are in close agreement with some slight differences that are negligible. Following
this, 3D printed springs are investigated and design guidelines are developed:
square wire cross-section is easier to print and mono directional in-fill produces
stronger springs. The springs are tested both in the pre and post-deformation
regions of the force-displacement plots. Testing showed that springs still behave
semi-elastically while after plastic deformation. Comparison between ULTEM
9085 and PLA springs lead to the observation that smaller layers increase the
stiffness of the springs. The next step of the research is the derivation of an equa-
tion for the calculation of the stiffness of 3D printed springs. Two methods are
attempted and one is successful, leading to an equation that predicted spring
constants that agreed very closely with experimental data. Finally a cost analysis
of springs printed with ULTEM 9085 is performed, PLA was not considered be-
cause it was only used for developing the model. Various methods for reducing
the cost are then investigated. The research presented in this thesis increases the
knowledge of 3D printed materials in several ways. First a characterization of
printed PLA is made which can be used as reference, given the printing settings.
The behaviour of 3D printed springs in the plastic region can be used as a safety
feature. Finally, the equation that predicts the spring constant can be used to save
time in design processes and is the starting point for developing more equations
that allow the full use of 3D printing’s design freedom for springs.
 CHAPTER 1

INTRODUCTION

1.1 Background

Additive Manufacturing (AM), also referred to colloquially as 3D printing, is a

novel manufacturing technique that involves depositing layers of material one

on top of the other to produce objects [46]. It is called additive manufacturing

to distinguish it from machining, joining, and forming which instead typically

remove material until the part is formed.

3D printing started as rapid prototyping in the 1908s where an early form of

Stereolithography (SLA) [47] for producing plastic prototypes was developed to

help visualization of parts during development. Today’s 3D printers are able to

produce functional parts as well as prototypes and many types of AM have been

developed, for example material extrusion [26], Selective Laser Melting (SLM)

[48], Selective Laser Sintering (SLS) [49], Digital Light Processing (DLP) [50], and

Rapid Freeze Prototyping (RFP) [51]. They all use the same principle of layering

material but differ in technique and material used.

AM is especially relevant in the production of complex and customized struc-

1
tures that used to be hard if not impossible to make [52, 53]. The space industry

is a sector where low volume, highly customised parts are the norm so it is per-

fect for AM. Many space agencies and private companies have begun using AM

for two main reasons. Firstly, mass savings from 40% to 90% are possible [54]

and mass is directly related to cost since heavier objects cost more to launch.

Secondly, AM can manufacture complex parts much faster than traditional man-

ufacturing, reducing fabrication times from one year to only 4 months [55].

Research in Additive Manufacturing For Space (AMFS), AM that is directly

applied to space, is varied. From using it to print CubeSat propulsion systems

[9], to printing ceramics [56], to using Laser Engineered Net Shaping (LENS) [57]

with Lunar and Martian regolith [58], to potentially printing an entire spacecraft

in orbit [59]. Regolith is the term used to describe the lunar soil [60] but can be

used to also for the Martian soil.

There are several problems with AM, though, since it is a manufacturing pro-

cess that is difficult to model. Despite this, progress has been made in character-

izing AM materials and efforts are being made to model and predict the material

properties [46].

Spacecraft, like any machine, require mechanical components to work. One

of the simplest and most used components is the mechanical spring, a device

that can store and release mechanical energy. They come in many shapes and

sizes and are used in many mechanisms, from clocks to vehicle suspension to

antennas on spacecraft [61]. One area that is lacking research is 3D printing of

springs, with an end goal to use them on spacecraft, and that is what this research

attempts to remedy.

1.2 Motivation

There are two main reasons for using 3D printing in the space industry: design

freedom and mass savings, which are intertwined. The latter is a consequence of

2
the former and is very important because mass is directly related to cost when it

comes to spacecraft, which is discussed more in subsection 2.2.1. At the moment

3D printed parts are mostly used in static applications, such as struts. This type

of loading also applies to some mechanisms and the idea of leveraging the design

freedom of AM for fabricating ad-hoc components for more dynamic applica-

tions is beginning to be considered a reality, there is an example of a mechanism

being developed for deployment of satellite solar panels that will be discussed

later. The lack of understanding of how springs produced via AM behave when

subjected to loading is the main driver for this research, in an effort to advance

the understanding mainly for the benefit of the space industry.

During the research, a phenomenon was also observed and further investi-

gated, which is the presence of localized strain oscillations observed during slow

tensile loading. Since this phenomenon lies outside the scope of this research,

the chapter about strain oscillations can be found in Appendix A.

1.3 Scope and objectives

The main focus of this research was to model the behaviour of 3D printed springs

for use in mechanisms employed on spacecraft. They can be used in, for ex-

ample, novel Cold Gas Thruster (CGT) designs or telescopic antennae and AM

lends itself particularly well to these types of applications since the springs can

be designed to be of any shape or size. The main material used was Polylac-

tic Acid (PLA) printed with a Fused Filament Fabrication (FFF) printer because

both are cheap, readily accessible, and printed PLA has a high degree of material

anisotropy, making the results of this research easily applicable to other material

types. A comparison was also done with springs printed with ULTEM 9085, a

high performance thermopolymer that has been used in space-flight already. As

mentioned before, an unusual phenomena was also observed and explored in

the course of the research, the observations made can be found in the appendix.

3
 A series of five sub-objectives was employed to achieve the main objective,

listed below.

1. Build a model for use in simulations and for reference. This was done via:

• identification of the most appropriate model for 3D printed materials

• determination of the most accurate way of measuring the strain dur-

ing testing

• measurement of the material constants that were required for the model

• comparison of the material constants based on the type of in-fill

2. Variation of the number of perimeters for samples subjected to tensile loading,

achieved via:

• simulation of tensile coupons with various numbers of perimeters and

the most commonly used in-fill

• printing and testing of tensile coupons with the same characteristics

• comparison of the simulations and experimental results

3. Using a novel thruster design or deployable helical antennae as possible

applications, investigate the performance of 3D printed springs by:

• determining the spring wire cross-section that will print best

• printing and testing a variety of helical springs, both compression and

tension

• testing the springs to explore differences and similarities between in-

fill and types

• perform cyclical loading in the post-deformation region

4. To reduce the time needed for design and testing, derive an equation that

calculates the spring constant of 3D printed springs:

• identify a derivation method

4
 • perform the derivation

• compare the calculated and experimentally measured spring constants

5. Develop methods for reducing the cost of printing springs by:

• determining the material/support ratio of springs printed with UL-

TEM 9085

• calculating the manufacturing time and cost of single and batches of

springs

• leveraging the design freedom of AM to reduce the cost

1.4 Structure

The next chapter, chapter 2, will review the research that has been currently done

on 3D printed springs and in AMFS in an effort to highlight the knowledge gaps.

Following this, chapter 3 will discuss in more detail how the objectives will be

achieved. Then chapter 4 will cover the methodology for measuring the ma-

terial constants for PLA, the plastic used in this research. The simulation and

physical testing of samples with varying numbers of perimeters is explained in

chapter 5. This is followed by chapter 6 where the findings of tests done on com-

pression and tension springs are discussed. Chapter 7 describes the derivation

of an equation for predicting the spring constant of 3D printed springs. The final

of the main chapters is concerned with costs of 3D printed springs, manufactur-

ing time, and methods for reducing the cost. This thesis will then conclude with

chapter 9, conclusions and future work. Each of the chapters apart from this

and the last one will have a short summary at the end. Appendix A is about the

exploration of localised strain oscillations that have been observed.

5
 CHAPTER 2

LITERATURE REVIEW: 3D
PRINTING FOR THE SPACE
SECTOR

This chapter will discuss the research and developments that has been made

in 3D printed springs and AM for use in the space industry. For AMFS, the

discussion is going to be specifically about the accomplishments and gaps in

AM research whose end goal is use in some aspect of space related applications,

from launchers to habitat building on extraterrestrial planets.

2.1 3D printed springs

There is very little literature that applies AM to mechanical springs. Most au-

thors make use of the following equation,

Gd4
k= (2.1)
8D3 Na

6
where k, the stiffness of the spring, can be calculated from the spring parameters

shown in Figure 6.1 and the shear modulus of the material, G. This equation

will be discussed in more detail in chapter 7 but has been stated here to empha-

size that most research assumes that 3D printed springs are made of isotropic

material, hence there is only one shear modulus.

In Saleh and Ragab [1], the authors studied the effects of geometry on the

shear modulus of helical springs manufactured with Ti-6Al-4V via SLM. They

solved Equation 2.1 for G then printed various springs with different pitches,

examples shown in Figure 2.1, to verify if changes in the geometry of the material

affected the shear modulus. The results revealed that the shear modulus ranges

from 60 to 85% of the rated value (38 ± 3.8 GPa) and that springs with a smaller

pitch were closer to the rated shear modulus. Interestingly the springs show no

signs of supports, likely due to their small size (42.04 mm in length and 22.14

mm in diameter), something which could be a problem with bigger springs.

Figure 2.1: Springs with four different pitches, four samples were manufactured per pitch value
[1].

He et al. created an interactive tool that allows users to create and control

3D printed deformable objects with embedded springs and joints [2]. First they

tested tensile samples made of tough PLA [62] to obtain the material properties.

Like Saleh and Ragab, they used Equation 2.1 to calculate the shear modulus

of the material and verified that it gave them good results when varying spring

parameters, see Figure 2.2b.

Hoa used 4D printing to manufacture and test leaf springs [3]. 4D printing

is a manufacturing method that combines 3D printing together with the ability

7
 (a) The Ondulé spring design tool interface [2].

(b) Results showing that 3D-printed helical springs have similar twisting performance to theoretical
predictions with varied d, D, N, and L values [2].

of the part to reconfigure itself upon activation of some mechanism such as heat,

light, or the absorption of moisture. The author used this manufacturing method

to produce curved leaf springs using flat moulds and the results showed that it is

possible to make composite springs with practical stiffnesses, like the one shown

in Figure 2.3, and strengths comparable with composite springs that are currently

used.

There is also some literature on composite material springs dealing with opti-

mization of spring performance [63, 64], substitution of metal springs with com-

posite ones [65], and tubular helical spring whose structure is made from multi-

ple laminae of composite material [66]. None of these works are relevant to this

research, though, because the manufacturing processes are very different from

AM.

Other research that has been found is on the application of 3D printing to

compliant mechanisms, which are flexible mechanisms that transfer an input

8
Figure 2.3: Example of a leaf spring created using 4D printing. When first produced the spring
was flat but it became curved when the right conditions were met [3].

force and displacement from one point to an output force and displacement at

another point through elastic body deformation. Only one source was found

that discussed 3D printed mechanisms and their application to space: a primer

produced by Merriam et al. on 3D-printed titanium compliant mechanisms for

aerospace applications made with Electron Beam Melting (EBM) [67]. There is

other research on compliant mechanisms outside of the space industry [68, 69,

70, 71, 72] but it is not relevant to this research since the focus is specifically on

springs.

The work by Saleh and Ragab, He et al., and Hoa were the only instances of

published literature that could be found that made use of AM to produce springs.

None of the studies had as an objective the in-depth investigation of 3D printed

springs, though, and made simplifications such as assuming that only one shear

modulus was necessary, implying that 3D printed material is isotropic. This is an

oversimplification of the material properties that will be discussed in more depth

in subsection 4.1.3. Due to the sparsity of published literature on the subject,

the conclusion can be made that there is a definite research gap when it comes

to the application of AM to 3D printing. Only simple models are proposed in

the literature and no detailed study has been made that develops guidelines for

9
springs manufactured via AM.

2.2 Research in AM for space

An attempt has been made to make this literature review as comprehensive as

possible, but it is very difficult to cover all AMFS because to the sparsity of details

when it comes to research in this specific area due to various reasons. Firstly gov-

ernmental agencies like National Aeronautics and Space Administration (NASA)

and European Space Agency (ESA) are at times not able to release sensitive in-

formation to the public so cannot go into much detail. Then there is the fact that

private companies such as Space-X, Boeing, Thales Alenia Space (TAS), and Air-

bus usually release even less details, most research news coming in the form of

press releases and finished products whose details are protected by intellectual

copyrights. Finally, there are relatively few research institutions such as universi-

ties that specialize in AMFS, most of the research is in general AM. More specific

review papers on general AM can be found for example on processes [73, 57],

materials [74, 75], simulation and modelling [76, 46], and cost models [77, 78].

2.2.1 The importance of mass reductions for spacecraft and launchers

Spacecraft are composed of many materials including various types of plastics,

metals and alloys, ceramics, composites, and glass [79]. Some, like plastics, do

not weigh much but metals such as aluminium, steel, and inconel that are used

for large parts of spacecraft are the main source of weight. As previously men-

tioned, one of the main ways that 3D Printing (3DP) can substantially help to

reduce costs is making parts lighter, which comes from the manufacturing pro-

cesses’ inherent design freedom. This short section discusses in simplified terms

the main way that 3D printing reduces costs and aids the space industry. An

10
example spacecraft moving in a circular orbit with speed v has velocity

r
GM
v= (2.2)
r

where G = 6.67 × 10−11 m3 kg−1 s−2 is the universal gravitational constant, M is

the mass of the planet being orbited, and r is the radius of the circle. For the

spacecraft to stay in a stable orbit, v is the velocity required.

Getting a spacecraft into space or altering its orbit requires changing its ve-

locity. When going from the surface of the Earth to space a launcher is used to

achieve this. The equation that governs the basic operation of any spacecraft is

called the Tsiolkovsky equation or the ideal rocket equation,

 
m0
∆v = ve ln (2.3)
mf

It relates ∆v to the exhaust velocity ve of the spacecraft’s thruster (the velocity at

which the fuel is being ejected), the inital mass m0 (which includes the spacecraft

and the fuel before the burn), and the final mass m f (which is smaller because

fuel has been used up during the burn). Equation 2.3 is simplified in ways that

are sufficient for the purposes of this thesis. It can be rewritten as

∆v
 
m0
= exp (2.4)
mf ve

m0
where is called the mass ratio. The plot of Equation 2.4 is shown in Figure 2.4,
mf
which illustrates that the greater the mass ratio, the greater the change in velocity.

If the mass of the spacecraft is small, the mass ratio increases and therefore the ∆v

budget increases for a given amount of fuel and the overall cost of the mission is

lowered. Lighter spacecraft have bigger ∆v budgets and therefore longer ranges

and/or lifetimes.

11
 40

m0
mf
20

0
0 1 2 3 4
∆v
ve

Figure 2.4: Final velocities as a function of launch vehicle mass ratios calculated using equation
Equation 2.4.

2.2.2 Usage of new technologies in the space sector

The National Research Council (NRC) summarises the ways in which the space

industry can benefit from AM in the following three points [59, 80]:

• The creation of new materials and parts that may one day be made in mi-

crogravity only and may only function there, i.e. they would be truly space

based

• Given the previous point, there would be a shift in the logistic and planning

of space missions since in-space construction would be a reality

• Lastly, given the previous two points, there is the possibility that AM could

change the space market by allowing construction of spacecraft directly in

space

The space industry is very conservative so new technologies are introduced

very slowly. This happens for two main reasons. Firstly, a long approval process

means that many years may pass until a mission’s actual implementation. For

example the Origins, Spectral Interpretation, Resource Identification, Security,

12
Regolith Explorer (OSIRIS-REx) mission was approved in 2011 and launched by

NASA in 2016 [81]. Considering the high rate of development of new technolo-

gies, especially in a fertile field such as AM, the time gap from the approval to

launch of the mission means that it is difficult to implement any new technolo-

gies or developments on the OSIRIS-REx spacecraft made in the interim.

Secondly, space missions are very expensive. Table 2.1 shows the rough costs

of launching to Low Earth Orbit (LEO) using SpaceX’s Falcon Heavy (FH) and

NASA’s Space Launch System.

Table 2.1: Cost to launch to LEO using the Space Launch System and FH [39], all costs are in
USD.

Launch vehicle Mass to LEO Cost to LEO Cost/tonne Cost/kg

Space Launch System 70 tonnes 600M 8.67M 8570
FH 53 tonnes 158M 2.97M 2970

Several costs are omitted from Table 2.1 such as development, construction,

testing, and certification of the payload before the launch. This can bring the

cost to 290 million USD for a satellite that can track hurricanes [82]. Constant

monitoring is also required once the spacecraft is in orbit which means further

expenses in the order of millions of USD per year for communication satellites

[82]. Given these high costs, and in some cases potential for the loss of human

life, it is no surprise that many entities in the space industry prefer to rely on

tried and tested methods and parts.

But AM has huge potential so there is a strong push to investigated this print-

ing technology for use in the space industry. For example, efforts have been

made to use 3DP to reduce the mass, complexity, part count, and welds of parts

[54, 52, 83, 80]. 3DP is now being considered of paramount importance for hu-

man exploration of space and the space sector in general [54, 84, 85].

13
2.2.3 In-space AM

Currently there is only one way to print in space, using the FFF printer on the

International Space Station (ISS), therefore most AMFS is still ground based. AM

research conducted in orbit will be referred to as In-Space Additive Manufactur-

ing (ISAM), which is a part of but separate from In-Space Manufacturing (ISM).

The latter includes also more traditional forms of manufacturing. ISM and ISAM

are closely connected therefore this section discusses both. Most of the ISAM re-

search has been performed as part of NASA’s In Space Manufacturing Initiative

[43].

The first in-space 3D printer was an FFF developed by California based Made

In Space (MIS) as part of the 3D Printing in Zero-G Experiment (3DPrint) [86, 42].

It was launched in September 2014 and fit in the ISS’s Microgravity Science

Glovebox (MSG). Several calibration objects and test coupons were manufac-

tured, Figure 2.5 shows some of the objects printed at this time. The samples

were then sent back to Earth to NASA’s Marshall Space Flight Center (MSFC).

There their properties were investigated and compared to a set of identical ob-

jects printed using the flight printer prior to launch. Tests included visual in-

spection, mass and density calculation, x-ray and Computer Tomography (CT)

scanning, destructive mechanical testing (done following ASTM standards for

tensile [87], flexural [88], and compression [89] testing) the results of which are

shown in Table 2.3 [90, 4, 41, 91].

The printer launched as part of the 3DPrint project was only the beginning

since as of 2016 there is a new FFF on the ISS: the Additive Manufacturing Fa-

cility (AMF) [40] shown in Figure 2.6, the specifications for which can be seen

in Table 2.2. The new printer has an expanded selection of materials includ-

ing Acrylonitrile Butadiene Styrene (ABS), Polyethylene (PE), and Polycarbonate

(PC).

Just like with the previous printer, the AMF was calibrated by performing

14
 Figure 2.5: Some of the objects printed on the ISS [4].

Figure 2.6: The AMF mounted in an EXPRESS Rack Mid-Deck Locker in the ISS (Image
credit: NASA).

Table 2.2: AMF system specifications [40].

Attribute Value
Print volume 14x10x10cm, 1400 cm3
Material ABS, Green PE, PC
Resolution 0.1 - 0.44 mm
Height Resolution ≥ 75 microns
Maximum wall thickness 1 mm

15
tensile, compression, and flexural tests using ASTM guidelines [30, 89, 88]. For

consistency the ABS used in these tests was the same as that used for the first

printer. Table 2.3 shows the test results of the two generations of printers both in

microgravity and on the Earth.

Table 2.3: Results from tests on samples prepared in orbit (0 g) and on the ground (1 g) [41, 42].

Environment 0g 1g
Test type AMF 3DPrint AMF 3DPrint
Ultimate strength (MPa) 38 27.9 37.8 23.9
Elastic Modulus (GPa) Tensile 3.1 1.7 2.2 1.5
Ultimate strength (MPa) 51.1 38.5 52.9 51.4
Compressive
Elastic Modulus (GPa) 1.7 1.1 1.5 1.7
Ultimate strength (MPa) 58.9 45.1 62.3 35.9
Elastic Modulus (GPa) Flexural 1.9 1.7 2.1 1.4

Being a newer printer, the AMF produced better quality parts, which is an

expected result. Compared to the samples printed on the ground, the flight sam-

ples had worse compressive and flexural properties but were better in tension.

MIS also investigated the phase distribution of the samples and found no signif-

icant differences [42].

The ability to manufacture custom tools in space will be invaluable to as-

tronauts, as shown by the Multipurpose Precision Maintenance Tool (MPMT) in

Figure 2.7. It is a tool that allows wire cutting and bolt tightening all in one and

was the winner of the “Future Engineers Space Tool Challenge” [5]. Thanks to

AM’s design freedom and the type of research exemplified by the AMF, astro-

nauts will be able to manufacture ad-hoc tools for any situation.

The first of only two references found that apply 3D printed springs to the

space sector is shown in Figure 2.8. These are a pair of small springs printed on

the ISS using the FFF printer there, called the AMF. The springs are made of ABS

but unfortunately no further information could be found on the dimensions or if

post-manufacturing testing was successful or not.

16
Figure 2.7: The MPMT, created by a university student as part of the “Future Engineers Space
Tool Challenge”, printed on the ISS [5].

Figure 2.8: Springs printed on the ISS [6].

The future of ISAM

The ultimate goal of ISM is to manufacture and assemble large structures in or-

bit. The obvious advantage of this is that it is far easier to launch raw materials

rather than pre-built delicate structures. Thus the cost of space missions would

therefore be lowered. An added benefit is that new spacecraft designs can be

developed because there will be no need to launch them [8]. There are many

challenges to implementing ISM though, these include:

• recycling of materials

• ability to move large structures during construction

• robotic dexterity in fabricating complex parts

• manufacturing of electronics

• being limited to materials that can resist the space environment

17
 The biggest issue is that currently there is no ISM infrastructure, meaning that

it would have to be built. This entails a large up-front cost that most agencies and

companies are not ready to pay for, therefore resorting to traditional Earth based

manufacturing [8], something that is discussed in more detail in subsection 2.2.2.

Closing the manufacturing cycle is also an important aspect of ISM. Just like

manufacturing on Earth, recycling of unwanted parts would help to reduce the

amount of raw resources needed. Recycling is especially important in manned

space exploration since it would lead to significant mass savings. ISAM, like any

other manufacturing technology, would also greatly benefit from this.

Tethers Unlimited, Inc. (TUI) has developed a “combination 3-D printer and

plastic recycler” called Refabricator to NASA for testing in flight [92]. Typically

when recycling polymer parts on Earth they are ground, which is not possible

on the ISS because the dust poses a health hazard to the astronauts. The Refabri-

cator instead uses a system called Positrusion, which does not involve grinding,

to recycle old polymer and print new parts[93]. The process has already been

demonstrated on Earth with FFF and ABS [94]. Interlog Corp., Techshot, and

TUI are also developing a system that can recycle multiple types of materials,

including metal and polymer [95]. NASA will then decide which company pro-

duced the best prototype and will continue the project with them.

Another important aspect of ISM is the ability to manufacture electronics.

Ultra-capacitors made with AM have already been developed, tested, and patented

[43] on Earth and the next step is to test this technology in flight.

An example of a solution to build large structures in space is TUI’s SpiderFab,

shown in Figure 2.9. It uses a combination of 3DP and on-orbit robot assembly to

create structures such as antenna reflectors, solar concentrators, solar sails, and

manned habitats. This achieved by a combination of two systems, Trusselator

and OrbWeaver. The former is a machine able to assemble large trusses through a

combination of AM and robotic assembly, while the latter is used to “manufacture

large, high-precision antenna reflectors on-orbit and then robotically integrate them with

18
a phased-array RF system” [96].

Figure 2.9: A concept art of SpiderFab Bot creating a truss in orbit [7].

Another project that is developing a similar technology is Dragonfly. This is

a system by Space Systems Loral for in-space reconfiguration and installation of

large RF reflectors [43]. MIS are also throwing their hat in the ring with Archin-

aut, a system similar to SpiderFab. It is a “free-flying space manufacturing and

assembly capability that enables advanced spacecraft and structures to be produced in

the space environment” [97]. Orbital ATK are developing the Commercial Infras-

tructure for Robotic Assembly and Services (CIRAS), another candidate system

of in-space hardware assembly technologies for performing ISM [98].

All the above technologies tend to focus on construction of large metal struc-

tures. McGuire et al. [8] instead proposed an architecture for a spacecraft equipped

with a large FFF to print polymers instead. The advantage is that plastic requires

less power than metal to print due to the lower temperatures involved. Met-

als also typically require post-processing in order to achieve their full properties

and this also requires a lot of power. The proposed spacecraft architecture uses

concentrated solar energy to melt the polymer. In order to ensure a consistent

source of light, the spacecraft would likely be in a sun-synchronous orbit, which

also helps to reduce the amount of temperature fluctuations and simplifies the

thermal systems.

19
 Figure 2.10: System architecture for in-space 3DP spacecraft [8].

2.2.4 Ground based AMFS

Gone are the days when AM used to be called rapid prototyping, nowadays it

is considered a valid manufacturing method for aerospace parts. In the many

facilities present on Earth, numerous techniques that allow the printing of an

increasingly wide range of materials [99] and of increasing complexity have been

developed. As a consequence, AMFS research can be broadly divided into three

categories based on the material used for printing: metal, polymer, and others.

These categories can be further divided into manned and unmanned according

to the type of mission that the printed parts would be used for. Research in

AMFS has been classified in this way in the following sections, an attempt has

been made to classify research under ‘manned’ only if it is specifically targeted

for this use. A brief description of satellite classification and what is special about

CubeSats follows.

Satellite classifications

Satellites come in many shapes and sizes but they are generally categorized

based on mass as follows [100]:

• Large: > 1000 kg

• Medium: 500 - 1000 kg

20
 • Mini: 100 - 500 kg

• Micro: 10 - 100 kg

• Nano: 1 - 10 kg

• Pico: 0.1 – 1 kg

• Femto: < 100 g

CubeSats are a particular type of nano-satellites that come in units of 10 ×

10 × 10 cm (1U) and can be configured into spacecraft of up to 12U in size, shown

in Figure 2.11. They are very small, simple, and cheap to make, so much so

that institutions such as universities [101] and even elementary schools [102] can

build and launch them. There has also been a case of an individual building

and launching his own CubeSat [103]. They are mostly built with commercial

off-the-shelf components and are launched in tandem with other CubeSats in the

fairings of bigger satellites, thus sharing the cost of launch.

Figure 2.11: Examples of different configurations of CubeSats [9]. Left - 1U “Phonesat”; top
right - 1.5U “EDSN Spacecraft”; bottom right - 6U “EcAMSat”.

21
Metal AMFS research

Manned missions AM can produce very particular geometries, which is one of

it is main advantages. In [104], the oxygen system study, a comparison was made

between wrought Inconel 718 and SLM printed In718 with post heat treatment

to determine the differences in material flammability for components printed for

the ISS’s urine processor assembly and oxygen systems. Fire is a major concern

on the ISS so the project’s goal is to investigate the flammability of printed parts.

It was found that printed In718 with heat treatment and Hot Isostatic Press-

ing (HIP) burnt more than printed In718 with only heat treatment and wrought

In718.

Unmanned missions Most AMFS research with metals is focused on propul-

sion systems using Powder Bed Fusion (PBF) techniques, especially SLM, since

this technology is reliable, among the most well understood, and most consis-

tent. Werkheiser [43] and Clinton [105] discuss an on-going project that NASA

has to develop a printed rocket engine prototype in 2.5 years by leveraging the

advantages of AM to reduce part count, cost, and fabrication times, further de-

tails can be found in Table 2.4. The developed engine prototype will then be

used as a basis for creating a new engine which will receive certification to use

on missions.

Carter et al. [106] presents a summary of the findings of NASA’s Glenn Re-

search Center (GRC) with regards to metal AM, which helps to give an overview

of the activities in this agency. Various fields were investigated and the findings

presented in the paper.

The first study took a replica RL-10 combustion chamber made of AM copper

alloy with the injector made of AM Ni alloy and subjected it to hot fire testing

[55]. 19 tests were performed in total with 4 different engine configurations and

the results were used to implement requirements for the AM parts. This work

also identified ways that AM’s design freedom could be used to improve perfor-

22
 Table 2.4: Developing a prototype rocket engine with AM versus conventional manufacturing
[43].

State of the art for Prototype AM engine

typical engine devel-
opments
Design, development, 7-10 years 2-4 years
test and evaluation
time
Hardware lead times 3-6 years 6 months
Testing Late in the DDTE cycle Early in the DDTE
cycle
Cost Engine: 20-50 million Prototype: 3-5 million
USD USD
Applicability Design specific to Provide relevant data
mission and often to multiple customers
proprietary and flexible test bed
can accommodate
changes in hardware
or design concepts

mance and reduce cost.

Another study mentioned by Carter et al. [106] is the creation of a material

database for Ti-6Al-4V printed with EBM [107, 108]. HIP was performed after

printing to close any porosities and to achieve the desired microstructure. De-

tailed analyses of the chemical and structural changes in the material were per-

formed throughout the process. High-Cycle Fatigue (HCF), CT, and tensile test-

ing were also performed. Ti-6Al-4V parts printed via EBM were found to posses

mechanical properties comparable or superior to conventional Ti-6Al-4V [109].

The database that was created for this material and process combination contains

details such as micro-structure, mechanical properties, fatigue crack growth, and

fracture toughness as well as thermal properties that were observed from cryo-

genic to elevated temperatures.

GRCop-48 is a copper alloy developed by GRC for rocket engine main com-

bustion chamber liners and has excellent mechanical and thermal properties [110].

Like EBM Ti-6Al-4V, GRCop-48 printed using Laser Beam Melting (LBM) [111]

23
has been found to produce parts whose properties are better than those made

with traditional means [112, 106].

Similar to the project discussed by Werkheiser [43] and Clinton [105], GRCop-

48 is also being used in the Low Cost Upper Stage-Class Propulsion (LCUSP)

project to reduce the cost and time taken to manufacture rocket engines [113].

Here two materials and two printing technologies were used together, GRCop-

84 was used to print a combustion chamber liner with SLM then Inconel 625

was used with Electron-Beam Freeform Fabrication (EBF3 ) [114] to manufacture

a structural jacket on the inside of the liner [108]. This was done in an effort to

develop processes for the rapid manufacture of reliable advanced engine parts.

EBM typically has one beam but a multi-agency team has investigated Ni-

based super-alloys using multi-beam EBM [108, 115]. The objective of this project

is to expand the capabilities of AM, specifically by developing alloys that lever-

age 3DP’s unique manufacturing methodology.

Large scale metal AM is one of the biggest areas of research worldwide due

to the fact that 3D printers are currently quite limited in size. NASA is also in-

vesting in this field with technologies such as LENS, EBF3 , Laser Freeform Man-

ufacturing Technology (LFMT), and various forms of Direct Energy Deposition

(DED) because, as shown in Figure 2.12, the build volumes of metal printers is

too small to be useful for large rocket engine manufacturing.

Not all research in large scale metal AMFS is done by governmental agen-

cies. Relativity Space is developing new launchers using the Stargate system, a

proprietary, automated, large scale metal 3D printing system developed by its

founders shown in Figure 2.13. Parts are manufactured thanks to robotic arms

that perform laser sintered metal printing [116]. Similar to other projects dis-

cussed, Relativity Space plans to simplify and reduce the cost of building launch-

ers by reducing the part count and development time. Terran 1, the company’s

first launch vehicle, is slated to have its inaugural launch in 2020 with a capacity

of 1,250 kg to LEO [117].

24
 Figure 2.12: Comparison of SLM build volumes to rocket engines [10]. Dimensions in SI
starting from the left are 25.4 × 25.4 × 25.4 cm (16387.064 cm3 ), 39.37 × 60.96 × 48.26 cm
(115823.768 cm3 ), 228.6 cm, 116.84 cm, 177.8 cm, and 142.24 cm.

Figure 2.13: Tim Ellis, co-founder of Relativity, holding a printed nozzle in front of the Stargate
printing system [11].

25
 Stepping away from large scale metal AMFS, Goddard Space Flight Center

(GSFC) has been conducting research into spacecraft instruments, components,

electronics, sensors, and coatings using technologies such as Direct Metal Laser

Sintering (DMLS) [57] and Atomic Layer Deposition (ALD) [118]. To manufac-

ture electronics, they have been making use of printers such as the Aerosol Jet

[80]. GRC (with Aerojet Rocketdyne) and MSFC have also 3D printed and tested

several injectors [80].

Unlike chemical propulsion rockets, whose energy source is the chemical re-

action that occurs in the combustion chamber, rockets that use Nuclear Thermal

Propulsion (NTP) employ a nuclear reaction to heat the fuel and then eject it out

though the nozzle, a schematic of which can be seen in Figure 2.14. It is an ex-

tremely efficient form of propulsion i.e. the mass ratio (defined in equation 2.4)

is very small. Mireles et al. [119] discusses the potential to use AM for NTP

because some of the parts needed are very difficult to produce otherwise.

Figure 2.14: Schematic of a NTP rocket [12]. Fuel is pumped into the combustion chamber where
a nuclear reactor provides the heat for combustion. This is much more efficient than traditional
rockets due to the far greater amount of thermal energy provided by the nuclear reactor.

In Europe the Additive Manufacturing Aiming Towards Zero Waste and Ef-

ficient Production of High-Tech Metal Parts (AMAZE) project (❝♦r❞✐s✳❡✉r♦♣❛✳

26
❡✉✴♣r♦❥❡❝t✴r❝♥✴✶✵✺✹✽✹❴❡♥✳❤t♠❧) was a multi-disciplinary effort to improve sev-
eral key areas of metal AM for the aerospace, space, energy, and automotive

sectors. It was the largest project of its type and had many objectives which

included increasing print volume and productivity, cost reduction compared to

traditional processes, improve dimensional accuracy, and reducing scrap rates

[120]. Two printing methods were investigated, PBF and DED performed with

powder and wire. The AMAZE project lead to many developments. The ones

that directly pertain to AMFS include [120, 121, 122]:

• Wire and powder feedstock specifications and test protocols to aid in de-

velopment of a robust supply chain

• Software for rapid and reliable design of 3D printed parts

• Build strategies and processes which enabled build times to be reduced by

a factor of 10

• Novel materials including aluminium alloys and composites which better

utilize the AM process

• New finishing procedures and flexible fixtures

• In-process monitoring methods to capture key process variables during

printing

• Improved process models

• A new benchmarking process, further discussed in section 2.3

• Many demonstration parts which have reduced cost, part count, weight,

and improved performance

Private companies also have a vested interest in metal AM. Apart from the

aforementioned Relativity Space, SpaceX has also been doing a lot of research in

AM which led to, for example, the launch of a Falcon 9 rocket with a 3D printed

27
Main Oxidizer Valve (MOV) [123]. Airbus Defence and Space used DMLS to

print the brackets shown in Figure 2.15 for their Eurostar E3000 telecommunica-

tions satellites [124]. NASA’s 2011 JUNO spacecraft also had printed brackets,

this time made by Lockheed Martin [121].

As mentioned before, one of the main research areas is the development

of topology optimization to create lightweight structures for space. An exam-

ple is shown in Figure 2.15, by Thales Alenia Space, who has launched 79 3D

printed parts spread across various missions, namely Telkom 3S, SGDC, and

KOREASAT-8 [121]. Other examples of this type of research include develop-

ing and using a process flow for additive manufacturing of topology-optimized

metallic components suitable for use on satellites [125, 126, 127, 128].

Figure 2.15: Titanium brackets manufactured using an EOSINT M 280 for use on Eurostar
E3000 satellites [13].

Universities also perform metal AM research and it sometimes extends to

AMFS. In Deepak et al. [52] the authors tested a novel liquid bi-propellant rocket

engine designed by the Students for the Exploration and Development of Space

at the University of California San Diego (SEDS UC) student association. Know-

ing of the design freedom granted by 3DP, the group used their ability to change

the design at any stage throughout the project. They were able to design a regen-

erative cooling system that was embedded in the combustion chamber’s walls

and has optimized channel cross-sections. The prototype rocket was printed us-

ing DMLS In718. The group designed a new version of their rocket and in 2018

28
they printed a liquid oxygen and kerosene engine again using In718 with DMLS

[129].

As previously mentioned, only two references were found that looked into

3D printed springs and the space industry. The first one used ABS printed on the

ISS while the second one uses metal. TAS has done some work on 3D printed

springs: they developed the Adel’Light hinge for deploying solar panels, shown

in Figure 2.16. The coil springs push against each other to generate torque and

open the solar panels and once in position the hinge locks in place. The prototype

shown in Figure 2.16 is made of titanium and would be very difficult to fabricate

without 3D printing. No published work could be found on this hinge.

Figure 2.16: The Adel’Light hinge for deployment of solar panels [14].

Polymer AMFS research

Manned missions As stated before, AM is extremely useful for long term hu-

man space exploration. They main advantage being that astronauts can manu-

facture ad-hoc tools when needed rather than having to carry them all the way

from Earth. In this spirit Wong et al. [85] carried out an investigation into the

quality of ABS surgical tools printed with FFF. Several surgeons were asked to

perform simulated prepping, draping, incision, and suturing using 3DP forceps,

hemostats, and clamps. The consensus was that the printed tools were adequate

even if they were printed with ABS. Nonetheless, the usefulness of AM for long

29
duration manned missions is greatly highlighted by this study.

Unmanned missions Injectors are an integral part of rocket engines and Catina

et al. [130] used FFF with ABS to print various configurations of injector plates

to test if an acceptable flow would be possible. The liquid used was water and

no hot fire testing was performed, which is understandable given the polymer

used. Although crude, the experiments showed that adequate flows could be ob-

tained. They also performed simulations of the injectors, compared them to the

real samples, and found that the two were quite different. This is to be expected

since there is a lack of software that is able to accurately represent materials pro-

duced with AM and will be further discussed in section 2.3.

The design freedom that AM has is again cast into the light by Marshall et al.

[9] discussing how AM could aid designers of CubeSats by allowing embedded

wiring, electronics, and propulsion systems. They incorporated a commercial

Busek micro Pulsed Plasma Thruster (PPT) into a structure printed with FFF.

Prior to this the team had tested whether printed PC, the material used, had a

high enough dielectric strength and results showed that the material could with-

stand the voltage without breaking down. The µPPT was fired while embedded

in the PC and the material again performed well as there were no signs of degra-

dation from the firing apart from some colouring on the casing due to arching.

Plastic is not as strong as metal so creating stronger polymers is an area of

research in AM. NASA had a study performed in this field where simple PLA

and PLA mixed with bronze, copper, iron, and stainless steel were compared

[131]. Tensile, wear, fracture, and microscopy testing was performed at different

layer heights and it was found that higher concentrations of metals lowered the

strength of the printed material but increased the stiffness and porosity while

Poisson’s ratio stayed about constant. What was likely happening is that the

plastic did not properly bond with the metal particles, thereby weakening the

material [131].

30
 Although not a written rule, CubeSats have been designed from the start to

be simple and cheap so a printed micro-thruster fabricated with AM for use on

nano-satellites should follow these guidelines. Gagne et al. [15] outlined the de-

velopment of a micro-thruster for use as either primary propulsion or for attitude

control, shown in Figure 2.17. The thruster respects the launch rules for Cube-

Sats [132] so only “green” propellants were considered. Printed with SLA using

Clear V2 resin, the thruster brought hydrogen peroxide (the mono-propellant)

into contact with different catalysts, resulting in an exothermic reaction whose

products are expelled through a converging-diverging nozzle. AM helped by al-

lowing the removal of a catalyst bed and the designing of an optimized nozzle

size.

Figure 2.17: Artist’s impression of the thruster in a 3U CubeSat [15].

Figure 2.18 shows the schematic of another proposed thruster design by Steven-

son et al. [16]. Unlike the one by Gagne et al., this one is designed for attitude

control of an interplanetary 6U CubeSat. Once again made with SLA, which pro-

duces the smoothest surface finish of all polymer printers, the printed portion of

the thruster includes the main propellant tank and plenum, seven nozzles, and

the propellant feed pipes. Fuel flow was regulated by solenoid valves and test-

ing in a thermal vacuum chamber showed that the thruster generated 50 to 60

mN of thrust with Isp =31.7 s and total impulse of 62.2 Ns [16].

Polymers can also be used to print CubeSat structures therefore Slejko et al.

[133] investigated the feasibility of printing modular, non-metallic CubeSat buses

31
 Figure 2.18: Diagram showing the tanks, pipes, nozzles, and valves for the thruster [16].

using Polyetheretherketone (PEEK) HP3, WINDFORM XT, and DuraForm PA.

PEEK turned out to be the strongest and since it has already been qualified for

use in space NASA [43], they decided to proceed with this material. The modu-

lar nature of the bus was achieved by printing and combining several elements

(bars, panels, etc.) to produce the size needed. A prototype 1U structure was

printed using SLS and is being tested to ascertain its mechanical properties.

There are also private companies that have an interest in developing poly-

mers for use in AMFS. For example Stratasys had partnered with NASA/Jet

Propulsion Laboratory (JPL) to print 30 polymer antenna array supports to be

used directly in space on the FormoSat-7/COSMIC-2 constellation [134].

Other AMFS research

Manned missions In-Situ Resource Utilization (ISRU) is the practice of using

locally sourced materials for construction and manufacturing. As humanity ex-

pands into space, ISRU will become increasingly important because it is much

32
cheaper and easier to build a habitat on Mars or the Moon than carrying one all

the way from Earth. NASA, along with the United States Army Corps of En-

gineers, was developing two sister modules for performing ISRU: Automated

Additive Construction System (ACES)-3 for terrestrial applications, shown in

Figure 2.19, and Additive Construction with Mobile Emplacement (ACME) for

extraterrestrial ones (using lunar or Martian regolith) [135].

Figure 2.19: Overview of ACES-3 [17].

When on Mars or the Moon, the best option is to use local regolith as aggre-

gates for cement when constructing buildings. ACES 3 has been used with lunar

(JSC-1A) and Martian (JSC Mars-1A) regolith simulants to make both normal

and “waterless” concretes. These concretes were then used with contour crafting

[136] to fabricate habitats to test the mechanical properties of the deposited ma-

terials, investigate various types of binders, and how the ACES coped with the

various concrete mixtures [17, 137, 138].

Another study tested four cement samples [17, 137, 44], three made with cast-

ing and one with additive construction, whose constituents are shown in Ta-

33
ble 2.5. Testing included simulated micrometeorite impacts, compression, and

visual inspection, which showed that cement with Lunar regolith had shallower

penetration depth than that with Martian regolith [44, 17].

Table 2.5: Constituents of the samples investigated in the ACME study [44].

Sample Simulant Cement Additives

1 JSC Mars-1A Portland Stucco mix
2 JSC Mars-1A Portland Stucco mix, Navitas 33
3 JSC Mars-1A Sorel Boric acid
4 JSC-1A Portland Stucco mix

Similar to the sun-powered FFF discussed in section 2.2.3, Meurisse et al. [18]

propose a sun-powered printer for making bricks on the Moon. This is an in-

teresting solution given the extremely thin atmosphere on the Moon and subse-

quent strength of solar illumination. As can can be seen in Figure 2.20 the printer

is basically an SLS that uses concentrated solar energy for printing instead of a

laser.

Results from 23 printed samples produced an average compressive strength

of 2.48765±0.71097 MPa and an average Young’s modulus of 0.20601±0.1519

GPa. The surface finsh of the samples was very rough and they were also very

porous. Given that there are still many limitations, this is a promising AM tech-

nique since it is being developed to be completely self-sufficient and does not

require powder sieving, making it ideal for ISRU.

Figure 2.20: Solar powered 3D printer with xenon lamps for testing [18].

34
 A study by ESA looked into the feasibility of using ISRU to build habitats

by using D-shaping [139] and Sorel cement [19]. Two developments were made

in the course of the study: the novel habitat design shown in Figure 2.21, and

a new lunar regolith simulant based on volcanic material. The habitat design

developed has a printed outer wall to offer protection against micrometeorites

while an inflated module inside provides atmosphere. Demonstration build-

ing blocks were printed to test the mechanical properties of the lattice and the

printer’s accuracy, both of which were found to be satisfactory.

Figure 2.21: Outpost structure (top) and wall profile (bottom left) with detail (bottom right)
[19].

It is not unfeasible that astronauts may have to use locally sourced materi-

als to make objects. A study used sieved lunar regolith simulant JSC-1AC in a

LENS printer to manufacture dense cylindrical parts and test their mechanical

properties, which were found to be satisfactory [58].

This work was later expanded by using SLM to study how scanning speed,

hatch spacing, laser power, beam diameter, and layer thickness affected the micro-

35
structure, surface roughness, and mechanical properties of various types of lunar

regolith samples manufactured via AM [140]. In this paper the authors also dis-

cussed how PBF printing processes would suffer in environments with reduced

gravity [140].

The studies discussed so far focus on solid objects whereas Jakus et al. [141]

instead investigated the feasibility of making a plastic-like material. They used

FFF to print inks made by combining sieved JSC MARS-1A and JSC-1A regoliths

mixed with elastomeric binders and a solvent mixture to make elastic materials

that were similar to polymers.

Asteroid mining is a potential source of vast amounts of resources [142] but

setting up the infrastructure is challenging. ISRU makes the task less daunting

therefore Lietaert et al. [143] used Direct Metal Printing (DMP) to print objects

using an iron meteorite as the source of the powder.

Unmanned missions As discussed in section 2.2.4, CubeSats are very small so

reducing the volume occupied by components is a must. AM’s design flexibility

would allow for great mass and volume savings so there are projects such as as

“3D Printing the Complete Cubesat” whose goal is to advance 3DP for CubeSat

applications [144]. Kief [144] also mentions that AM could enable the develop-

ment of new structures where several subsystems like power, propulsion, and

communications are incorporated directly into the structure. This would require

printing of various types of materials hence why this research has been put in

this subsection.

The final form of printing that does not involve metals or polymers is ceram-

ics and ceramic based composites. Because there is a wide variety if printing

techniques, there is the possibility of developing new materials that have unique

properties. For example ceramic parts with fibre reinforcements are considered

to be possible with Laminated Object Manufacturing (LOM) [145] and Binder

Jet Printing (BJP) [146]. Other AM techniques can be used though, and have

36
been demonstrated on gas turbine components using both ceramics and poly-

mers [147, 55].

There is also research in a more general field that is not necessarily material

specific: Design for Additive Manufacturing (DFAM) applied to the space sector.

Dordlofva et al. [148] proposed developing “Design for Qualification” guide-

lines for AM components by considering qualification as a design factor in the

early phases of product development, instead of more traditional approaches,

thus reducing cost and lead time for development and qualification as products

are designed to be qualified. Borgue et al. studied how the introduction of 3DP in

space applications impacts the design phases [149]. They found that most of the

literature agrees that there is need to better understand the constraints and be-

haviour of AM so as to take full advantage of its increased design freedom and

that there are few studies on systematic and generic modelling approaches for

DFAM for space, unlike in other industries. Also, a constraints replacement strat-

egy for the redesign of components to be manufactured with AM based on En-

hanced Function-Means modelling has been proposed as a basis for re-design of

products using AM applied on a case study featuring a satellite sub-component

[150, 151].

2.3 Research in AMFS and 3D printing of springs: gaps

and directions

One of the main problems, which affects all AM and not just the space sector,

is that there are many factors that determine characteristics and therefore the

behaviour of printed parts. Because of this, their interaction and effects still not

well understood. For example SLM can have as many as 130 different factors

affecting the quality and performance of the final object [152]. Many factors are

semi or fully random, such as powder distribution or the flow of melt pools,

making AM difficult to predict. Even two objects printed together may have

37
large feature variations [153] and this is a fact that cannot be avoided, simply

mitigated. Ghidini [54] summarises the challenges in widespread adoption of

AM in the space industry very well:

• the need to change the approach to design in order to take full advantage

of the capabilities of AM, in other words more research into DFAM;

• the need to ensure reproducibility, accuracy, and reliability;

• the need to develop new standards and verification methodologies for 3D

printed parts since classical standards do not apply, also mentioned by [53].

Most of the issues facing widespread adoption of AM can be mitigated by the

adoption of standards. Fortunately, there is a lot of activity in AMFS and slowly

standards are being introduced. outside of the space industry, ASTM Interna-

tional and ISO have published a structure for their future standards regarding

AM which are general and not specific to the space industry. ASTM F3122 [154]

provides guidelines on how to evaluate the mechanical properties of materials

made using AM and references several existing standards among which ASTM

A370-16 [155], ASTM B565-04(2015) [156], ASTM E132-04(2010) [157], and ASTM

E290-14 [158]. Similar to ASTM F3122 the America Makes & ANSI Additive

Manufacturing Standardization Collaborative (AMSC) is a cross-sector coordi-

nating body whose objective is to accelerate the development of industry-wide

additive manufacturing standards and specifications consistent with stakeholder

needs and thereby facilitate the growth of the additive manufacturing industry.

The AMSC published its Standardization Roadmap for Additive Manufacturing

(Version 2.0) in June 2018 [159]. This identifies existing standards and specifica-

tions, as well as those in development, assesses gaps, and makes recommenda-

tions for priority areas where there is a perceived need for additional standard-

ization. The AMSC roadmap describes 93 gaps [160] where no published stan-

dard or specification currently exists to respond to a particular industry need,

which have been compiled in an interactive, online portal [160].

38
 Space entities are also developing standards. NASA’s Additive Manufactur-

ing Structural Integrity Initiative (AMSII) project aims to establish a qualification

method for SLM parts to be used in space-flight applications. Two documents

have so far been produced: Engineering and Quality Standard for Additively

Manufactured Spaceflight Hardware (MSFC-STD-3716) [161] and Specification

for Control and Qualification of Laser Powder Bed Fusion Metallurgical Pro-

cesses (MSFC-STD-3717) [162]. The first contains the overall guidelines while

the second provides details for design process, part classification, pre-production

and production, manufacturing, qualification, and acceptance [163]. Together

MSFC-STD-3717 and MSFC-STD-3716 provide a framework for standardization

of laser PBF of metals.

As can be seen in this literature review, there is great interest in the space

industry for AM. Many areas have ongoing research including rockets, CubeSat

structures, surgical instruments, habitats, and electronics. Although the main

issue affecting AM in the space industry, and in other industries, is the lack of

established standards, for the purposes of this research this is not relevant.

There is a lot of work underway for the development of 3D printing stan-

dards for the space industry but, as previously mentioned, there is very little

literature about the application of 3D printing to springs, and what little there

is superficial. There are many potential applications for springs in spacecraft

mechanisms therefore the main objective of this research was the investigation

of springs manufactured via AM with the targeted application being, but not

limited to, use in the space sector.

Summary

This chapter discussed the literature reviewed for this research and the gaps.

While there is a lot of work being done in many areas of AMFS, there is very

little when it comes to 3D printed springs. Only three published papers were

39
found that investigated springs made via AM and the research performed in

them was superficial. Also, no published literature could be found that applies

AM to springs in spacecraft. Therefore a research gap has been identified and

this research will attempt to remedy it.

40
 CHAPTER 3

PROPOSED RESEARCH
METHOD

So far most AMFS has been used mainly for static parts such as combustion

chambers, piping, brackets, or buildings. But the design freedom afforded by

AM should also be considered when developing mechanisms like the aforemen-

tioned compliant mechanisms. Modelling parts for mechanisms will allow AM

to be used in unconventional ways, taking full advantage of its design freedom.

There are many types of springs, examples of which can be seen in Figure 3.1.

The most common is the helical spring, either compression or tension. Springs

are a complicated system to model and creating equations to predict their be-

haviours is not easy but they are widely used in mechanisms and therefore the

potential application of AM should be investigated. For this reason, a hypothet-

ical application of springs has been developed for this research, detailed in the

next section.

41
 Figure 3.1: Examples of types of springs [20].

3.1 Novel CGT design for small spacecraft

3.1.1 Cold gas thrusters

CGTs are simple thrusters used for attitude control in spacecraft and have been

hugely successful [164]. They work on the principle of ejecting pressurised gas

from a nozzle so, unlike in a chemical rocket, the gas pressure is not created via

a chemical reaction, it is simply stored under pressure. If the propellant tank

was directly connected to the nozzle the thrust would drop proportionally to the

drop in pressure due to the expulsion of propellant. By instead storing gas in

plena like in Figure 3.2, the pressure can be regulated and therefore the thrust

produced is more consistent.

3.1.2 Using AM to increase the dynamic range of CGTs

When CGTs are used in small spacecraft such as CubeSats, where space is a

valuable commodity, the inclusion of plena reduces the amount of propellant

available. They could be removed if there was a way to keep the propellant pres-

surised throughout most of the thruster’s lifetime. A hypothetical solution is

shown in Figure 3.3. This is a new CGT design that foregoes the plena and uses

42
 Figure 3.2: Example CGT module design schematic showing the main components: propellant
tank, plena, valves, and nozzle [21].

springs to keep the fuel under pressure. When the thruster is empty, the panel

rests on the left side and the springs are in a their rest configuration. As fuel

is introduced, the panel moves and pulls the springs, pressurising the gas. The

valve is then activated when needed and the fuel is expelled. As the propellant

is used the pressure drops but the force generated by the springs helps to keep it

constant.

Figure 3.3: A hypothetical CGT that uses 3D printed springs.

The thruster is assumed to fit in a CubeSat so has the following restrictions

[83]:

• dimension of 1U, see section 2.2.4

• mass: 1.3 kg per 1U

• the spacecraft is not allowed to have any pressure vessels, which NASA

defines as vessels that have pressures above 100 psia, which is about 689

43
 kPa

This means that the force generated by the springs will have to be tailored to

not exceed the tolerances or the thruster will be considered a pressure vessel and

thus will be subject to greater testing. The dimensions give a starting point for

sizing the thruster and therefore how much pressure the panel generates based

on the force of the springs.

3.1.3 The problem with linear springs

Cylindrical helical springs obey Hooke’s law,

F = kx (3.1)

which relates the force F needed to displace a spring from it is initial position by

an amount x via k which is called the spring constant or stiffness and has units

of N/m. When the spring is behaving according to Hooke’s law it is said to be

elastic. Figure 3.4 shows how the force and the displacement are defined for a

tension spring.

Figure 3.4: The spring at rest position is defined as x = 0. When a force is applied the
displacement from the initial position is called x.

As can be seen, Equation 3.1 is a linear equation so springs in the thruster

will produce less force the closer they come to their original position. In reality

springs can be stretched beyond their elastic limit but that will be discussed in

44
chapter 6. A solution is development of a non-linear spring, one who’s equation

is

F = f (x) (3.2)

where F is a function of x. In this case

dF
k= (3.3)
dx

Here the force is a function of displacement from the equilibrium position but

unlike in linear springs k is not a constant. For example when a compression

force is applied to the spring showed in Figure 3.5 the region with a smaller

pitch will compress first followed by the region with more pitch. This results in

a changing k and hence the name non-linear springs.

Figure 3.5: Example of a non-linear, dual pitch spring [22].

Although commonly used, non-linear springs have never been 3D printed.

There are design limitations due to manufacturing difficulties, like for example

it is not possible to vary the diameter or shape of the wire. This is instead not a

problem with AM due to the fact that objects are built layer by layer. Therefore

one of the objectives of this research was to derive an equation for the calculation

of the spring constant of 3D printed springs. The derivation technique could also

be extended to other types of springs such as spiral or conical. A series of linear

springs could then be created similar to Figure 3.5 to provide a force that is as

45
tailored to the requirements of the mission. For example if a CubeSat is expected

to tumble out of control after launch, a spring which initially generates a high

force can be manufactured to provide high thrust to stop the unwanted motion.

Once the spacecraft is under control, a the spring can be tailored to provide less

thrust for finer attitude adjustments.

3.2 Deployable helical antennae

Another example of a potential application of 3D printed springs is deployable

helical antennae. The predominant antenna choice for CubeSats has been one

or more monopole or dipole antennae, although they are low gain, naturally

linearly polarized, inefficient and normally narrowband [165]. There are many

types of deployable antennae available for CubeSats, shown in Figure 3.6a, but

the ones of interest here are the helical type. Helical antennae are wider band,

naturally circularly polarized, and may also have more than one helix to further

increase their gain [24]. Figure 3.6b shows how a helical antenna is folded and

stored.

Since helical antennae have the same shape as springs, it is possible, as will

be shown later, to print hollow polymer springs and then thread conductive into

them, even with several helices. There are two advantages to this method. First,

the springs act as the supporting helices (see Figure 3.6a) so there is no need for

dedicated support helices. Second, using the equation developed in this research

it is possible to tailor the stiffness of the spring so as to avoid unwanted oscilla-

tions as a result of deployment. It is also possible to create compound springs,

thereby providing a graduated extension.

3.3 Research method

Given the previous information and literature review the main goal of this re-

search was to model a 3D printed helical spring by developing an equation that

46
(a) Packaging schemes for CubeSat antennae: (a) the helical pantograph, (b) coilable conductors, (c)
dual-matrix composite shells, (d) hinged ribs, and (e) the wrapped mesh [23].

(b) Simulation of compaction of eight-helix pantograph antenna [24].

Figure 3.6

47
predicts their stiffness based on material properties and spring dimensions.

3.3.1 Material selection

The first step was determining which material to use for printing. This depended

on several factors including cost, printing flexibility (when it comes to the slicing

program), and ease of printing (because many samples of different types would

have to be manufactured).

Metal printing was not considered for the following reasons. First there was

the concern of support removal. Although Saleh and Ragab [1] showed that it

is possible to print small springs with no supports on an Elctro Optical Systems

(EOS) SLM printer, at the beginning of the research the need for supports for

larger springs was considered to be inevitable. Since they would also be printed

out of metal their removal would be very difficult and add a lot of time to the

process, thereby slowing the whole research. Removal of supports from springs

printed with metal could be investigated in a future study where techniques for

printing metal springs in a way that was efficient in both time and cost are inves-

tigated.

Another reason for not using metal is because of the increased mass com-

pared to polymers. Since the emphasis of this research was developing 3D printed

springs for use in the aerospace industry, as subsection 2.2.1 explained weight re-

duction is always a great consideration. Also, since the loads that spacecraft are

typically subjected to during flight are very small (being in microgravity) there

is little need to make springs out of metal.

The final reason is that 3D printed metals and plastics are both quite anisotropic

in their material properties [74, 166, 167, 168] therefore it makes little difference

if one or the other is used for testing. Out of the two, polymers tend to produce

materials that are more anisotropic therefore they were selected because they are

a “worst case scenario”.

The research is focused on creating a spring that is useable in space with

48
FFF, therefore the filament that was the primary candidate was ULTEM 9085 by

Stratasys, printed using Fused Deposition Modeling (FDM). This thermoplastic

and printing technology is certified for use in space since ULTEM 9085 was used

to manufacture the antenna array supports for COSMIC-2 [169]. Use of springs

printed with ULTEM for space applications would automatically qualify them

for flight. The only testing that would need to be performed is, for example, for

the effects of cyclical loading on the springs, or how the temperature changes

caused by the day-night cycle while in LEO affect the stiffness over time. These

tests are different for every application so are not covered in this research.

There are two issues with using this filament though. ULTEM 9085 is expen-

sive at 1320 SGD for a 1510 cm3 canister of material and 1320 SGD for a canister

of support, to which need to be added a set of tips (450 SGD) and print sheets

(1100 SGD for a packet of 20). Since this research includes a lot of experimen-

tal work, several canisters of ULTEM and support would have been needed so

a cheaper alternative was sought. The other reason for not using ULTEM 9085

for the bulk of testing was that the printer used, a Fortus 450mc, requires the use

of Stratasys’ proprietary slicing software. This does not allow manipulation of

layer height and temperature for ULTEM 9085, which ensures consistency but

does not allow flexibility. Therefore at the beginning of the research the deci-

sion was made to use a material/printer combination that allowed more print

settings to be changed.

Several plastics are available for FFF printers including PLA, ABS, PC, Ther-

moplastic Polyurethane (TPU), and Polyethylene Terephthalate Glycol (PETG).

PLA was selected because it is a commonly used polymer [170] (the other most

common one being ABS), it is easy to print [171], and has a higher rigidity, tensile

strength, and layer adhesion compared to ABS [172]. While not being certified

for use in space, PLA is cheap (at only a few tens of dollars per kg) and can be

printed on a variety of FFF printers. There are many slicing programs available

for PLA, all of which allow minute control of print settings, thus creating a larger

49
design space than with ULTEM 9085. This allowed the variation of layer height

that was tested in chapter 6. By using PLA on an entry-level printer more sam-

ples could be printed faster and with more flexibility in the print settings. The

equations derived in chapter 7 were tested with both PLA and ULTEM 9085 and

can be used with any material as long as the necessary shear moduli are pro-

vided. It should be noted that PLA was only used for developing the equation.

No intention is made to use PLA for functional space parts due to its high out-

gassing [173]. The material of choice for functional parts remains ULTEM 9085

since it has already been used in space.

3.3.2 Comparing springs made via AM and conventional means

The idea of plastic springs has been around since the middle of the 20th century

[174, 175] and they find use in niche applications such as extending the life of

hard disk drives by dampening vibrations [176], interferometers [177], and more

commonly in suspension systems [166] and applications where corrosion resis-

tance is necessary [178]. The materials used are usually PEEK, ABS, TPU, and

proprietary polymers such as special formulations of Polyetherimide (PEI). It is

therefore not possible to directly compare the springs printed in this research

with commercially available ones because none could be found that are made of

PLA or ULTEM 9085.

If the comparison were possible, due to the fact that print settings make a

large difference to the mechanical properties of printed parts it would be diffi-

cult to gauge if springs made via AM are consistently better, worse, or the same

without an in depth study. Future research could therefore be performed in this

field, comparing the properties of printed and conventional springs in an effort

to create printing guidelines that enable high quality printed springs to be man-

ufactured. The objective of this research, though, is not to compare printed and

existing springs but to enable new designs of springs to be manufactured via 3D

printing and therefore direct comparison with existing springs is not part of the

50
scope.

3.3.3 In-fill selection

After the material had been selected, the in-fills were considered. Figure 3.7

shows the cross-section of an object printed with material extrusion, where two

separate regions of material can be seen. The first is the contours or outline,

which are the tracks of material deposited on the outside of the part to create

smoother surfaces. Contour lines are always parallel to each other over all lay-

ers and follow the outline of the object. The contour width usually cannot be

changed but the number of outlines Nout can, with the standard in most slicing

programs being 2 or 3.

Figure 3.7: Cross-section of a part printed with FFF showing various features.

The other region is the in-fill, which is everything inside the contour. The

in-fill is typically produced by laying parallel tracks of material in alternating

directions so that one layer has raster angle (as defined in Figure 3.7) +45 and the

next -45. Theoretically there are infinite types of in-fill patterns (Archimedean

spiral, gyroid, grid, honeycomb, Hilbert curve, etc...) and directions (±45, 30/60,

0/90, 0/45/90, etc...) depending on the slicer used but most have ±45 as their

default setting.

Due to the different alignments, the in-fill and the perimeter can be defined

as two different materials. That is because the perimeter is composed of mono-

directional tracks that are parallel to the edges of the part while the in-fill is more

similar to a sandwich of alternating fibres like a composite. This difference is

very clear in Figure 3.7.

51
 Users typically measure the mechanical properties of test objects printed with

standard settings defined by the slicing program. They then measure the me-

chanical properties of those objects and perform simulations based on the recorded

properties. The perimeter, being different materials, could have an effect on the

mechanical properties of the part depending on how many there are. Two sets

of PLA samples were therefore printed, one with in-fill in the +45/-45 directions

(or ±45) and the other with only +45 (or mono-directional, also shortened to mono

in this thesis). These were used to simulate a tensile coupon using Finite Element

Analysis (FEA). Since the effects of raster angle has already been extensively in-

vestigated in the literature [27, 166, 179, 180, 181, 167, 182], there are no further

contributions that can be made in this research. Therefore the ±45 in-fill was se-

lected for the simulations because, as previously stated, it is the standard in-fill

in most slicing programs.

Apart from the FEA simulations, the two in-fills were selected for other rea-

sons. Adoption of the mono-directional in-fill is related to the mathematics of

creating an equation that predicts the stiffness of printed springs. The derivation

process will be discussed in more detail in chapter 7, for the moment it is enough

to say that an in-fill had to be selected for the springs that was constant along the

coils so as to not overcomplicate the problem. The ±45 in-fill varies as shown in

Figure 3.8a, at position 1 the in-fill has one configuration while at position 2 it is

different. Logically this means that the shear moduli would change with the in-

fill, making them dependent on the angle at which the cross-section is taken. All

in-fills that are deposited in straight lines, like 30/60 or 0/90, would cause the

shear moduli to change along a coil, making the derivation more complex. Fig-

ure 3.8b shows that the mono in-fill does not suffer from this problem because it

is consistent throughout the coil, hence why it was the main in-fill for the springs

manufactured in in this research.

52
 (a) Cross-section ±45 in-fill (b) Mono-directional infill has a
changes along the coil therefore constant cross-section
the shear constants do so too. throughout the coil.

Figure 3.8: Comparison of ±45 and mono-directional in-fills.

3.3.4 Procedure

After selection of the material and in-fills, the model that was best suited for 3D

printed objects was determined. Then all the constants needed for Equation 4.11

were measured for PLA using the two types in-fills. This was done in order to

better understand the material and for future simulations and modelling. It was

during this part of the research that the localised strain oscillations were first

observed thanks to the particular strain measuring technique employed. More

details can be found in Appendix A.

Once the constants had been measured the simulations of samples with vari-

ous numbers of perimeters was performed, followed by testing of samples with

the same number of perimeters. Most users of AM use pre-determined print-

ing parameters with little care for the effects of variables such as the number

of perimeters because they are typically interested in printing high quality and

consistent parts. The perimeters and in-fill can be considered separate materi-

als though, due to the way they are printed, therefore changing the number of

perimeters might have an effect on the tensile properties of the part. Simulating

the material as orthotropic with 1 to 5 perimeters and with different materials for

the in-fill and perimeters then verifying the results experimentally allowed this

question to be answered.

53
 In order to derive an equation for 3D printed springs, samples had to be man-

ufactured to determine if the calculated values corresponded to reality, there-

fore in this part of the research several springs were printed and their spring

rates measured. First a comparison of the printing quality of round and square

springs was made. Then three spring characteristics were varied: the pitch, coil

diameter, and wire diameter. k, the spring stiffness, was measured for all the

springs printed, both compression and tension, and for the different types of in-

fill. Cyclical testing was also performed on springs that had been stretched past

plastic deformation and into the second region to test their response.

Lastly, once all the data had been gathered, derivation of the equation for

calculating k was done. This involved the creation of a new formula since, as

explained earlier, 3D printed springs are very unusual. In fact the only equations

known are for isotropic materials, not even composites. The energy stored by

a stretched spring was equated to the energy stored by a shaft of equal length

subjected to torsion. A correction was included to account for the fact that the

equivalent shaft does not have a constant cross-section then the spring constants

were calculated and compared to the measured ones. Finally, the cost of printing

springs with ULTEM 9085 was calculated and some methods were developed for

reducing the amount of material needed, print time, and therefore cost.

Figure 3.9 shows a summary of the main objective and the sub-objectives

presented in section 1.3. For each step of the sub-objectives the section and page

numbers where the relevant details can be found are provided.

Summary

In this chapter, although ULTEM 9085 has been used in space already, due to the

cost it was determined that PLA was better for this research. Two in-fills, ±45

and mono (+45), were then chosen for testing. Finally the testing procedure was

outlined, involving first choosing a material model, then measuring the material

54
 4.1 (pg 57) 4.3 (pg 69) 4.4 (pg 79)

5.2 (pg 88) 5.3 (pg 89) 5.4 (pg 91)

6.2 (pg 95) 6.3.1 (pg 108) 6.3.2 (pg 122)

7.2 (pg 136) 7.4 (pg 147), 7.3 (pg 139) 7.4 (pg 147)

8.1 (pg 161) 8.2 (pg 162) 8.3 (pg 164)

Figure 3.9: Thesis roadmap.

constants, followed by comparison of simulated and experimental tensile testing

with varying numbers of outlines. Then 3D printed springs are to be investigated

in depth, and finally two equations for calculating the spring stiffness are to be

derived and verified.

55
 CHAPTER 4

CREATION OF AN
ORTHOTROPIC MODEL FOR
PLA

The objectives for this chapter was the following: build a model for use in simu-

lations and for reference. This was done via:

• identification of the most appropriate model for 3D printed materials

• determination of the most accurate way of measuring the strain during

testing

• measurement of the material constants that were required for the model

• comparison of the material constants based on the type of in-fill

56
4.1 Determination of material model

4.1.1 Reference axis and coordinate origin

In this research the reference axes for the printer (shown in Figure 4.1) are defined

following ASTM standards [183]:

• Z-axis: runs normal to the build platform;

• X-axis: runs parallel to the front of the machine and perpendicular to the

Z-axis;

• Y-axis: perpendicular to both X and Z axis so the right hand rule is satis-

fied.

The origin of the coordinate system is defined as the point where all three of the

coordinate systems intersect. Unless otherwise stated, this is assumed to be the

centre of the build platform.

Figure 4.1: Definition of printer axes and coordinate origin [25].

4.1.2 Road angle

The road angle, also called raster angle for FDM, is defined as the angle the

printer makes with respect to the X-axis in the bottom layer while looking along

the -Z direction, shown in Figure 4.2.

57
 Figure 4.2: Cross-section of a part printed with FFF showing various features including road
angle [26].

4.1.3 Anisotropic materials

In order to create a model for 3D printed springs it is necessary to first discuss

the nature of materials fabricated via AM. A short discussion of the mathematical

models used for anisotropic materials will now be given. Stress (compressive or

tensile) is defined as the amount of force F exerted over an area A,

F
σ= (4.1)
A

Strain (compressive or tensile) is typically a measure of the deformation of an

object where L0 is the initial unstressed length and L f is the length after stress

has been applied,

L f − L0 ∆L
ε eng = = (4.2)
L0 L0

ε eng is known engineering strain and is used in most applications. Logarithmic or

true strain was instead adopted in this research due to the ductile nature of PLA

and because it is required by Abaqus CAE [184], the FEA software used. The
dL
derivation of true strain begins with an infinitesimal strain increase dε t = L .

The true strain is then obtained by integrating from L0 to L f

Z εt Z Lf
1
dε t = dL (4.3a)
0 L0L
Lf
 
ε t = ln (4.3b)
L0

58
 Stress and strain are related by a form of Hooke’s law:

σ = Eε (4.4)

where E is Young’s modulus or the modulus of elasticity. Poisson’s Ratio (ν) is

the ratio of transversal strain (ε trans ) to axial strain (ε axial ) in response to stress

applied in the axial direction,

ε trans
ν=− (4.5)
ε axial

Referring to Figure 4.3, the shear modulus or modulus of rigidity (G), is de-

fined as the ratio of shear stress τxy (the component of σ that is coplanar to A) to

the shear strain γxy (the component of ε parallel to A),

τxy Fl
G= = (4.6)
γxy A∆x

∆x
Here F is a force applied parallel to and area A, γxy = l = tan(θ ) is the shear
strain, ∆x is the transverse displacement, and l is the initial length.

Figure 4.3: Definition of shear strain.

E, ν, and G are related in an isotropic material because the molecules in-

side are uniformly distributed, allowing the calculation of one property from

the other two. Domingo et al. [185] suggested an equation for calculating the

in-plane shear modulus,

Ei
Gij = (1 + νij ) (4.7)
2

59
where i is the loading direction and j is perpendicular to i. To calculate the shear

modulus using this method a type I tensile coupon is printed in the ij plane

oriented in the i direction but rotated by 45◦ . This method was not used here

because not enough literature could be found that confirmed if the method pro-

duced reliable results; so it was deemed more accurate to measure the shear con-

stants directly. A future study could be made where Domingo et al.’s method is

compared to the results obtained using ASTM standards.

3D printed materials, especially those produced via material extrusion (which

is the printing method used for this research), act anisotropically for of three rea-

sons [27]. First, during printing the molecules in the plastic align themselves

to the direction of deposition, thus responding to stress differently depending on

the direction. Secondly, this is also true of inter-layer bonding compared to bond-

ing between tracks, where the former is stronger than the latter. Finally gaps and

porosities can form between tracks like in Figure 4.4, creating areas where the

stress response is dissimilar from that of the surrounding plastic.

(a) Scanning Electron Microscope (SEM) (b) Cross-section of material extrusion

image of material extrusion printed plastic printed ABS that shows the air gaps
that shows the gaps between tracks [27]. produced during the printing process [28].

Figure 4.4

When tested AM materials will produce a standard stress-strain curve like

the one in Figure 4.5. Only the elastic portion of the curves is considered for

measuring the material properties. When in this regime, the generalized form of

60
 Figure 4.5: Typical stress-strain curve for a material. From A to B the material obeys Hooke’s
law (elastic behaviour). At C it starts to yield, from C to D it undergoes necking, and finally at
E the material breaks.

Equation 4.4 for a material undergoing elastic transformation is

    
σ1  C11 C12 C13 C14 C15 C16  ε 1 
    
σ  C C22 C23 C24 C25 C26 
 ε 2 
 
 2   21
    
σ3  C31 C32 C33 C34 C35 C36  ε 3 
    
 =   (4.8)
    
σ4  C41 C42 C43 C44 C45 C46 
 ε 4 
 
  
    
σ  C C52 C53 C54 C55 C56 
 ε 5 
 
 5   51
    
σ6 C61 C62 C63 C64 C65 C66 ε6

where the Cij constants form the the stiffness matrix and can be determined ex-

perimentally. Solving Equation 4.8 for the strains gives the following compliance

matrix where the Sij s are known as compliances,

    
ε 1  S11 S12 S13 S14 S15 S16  σ1 
    
ε  
 2  S22 S23 S24 S25 S26 
 σ2 
 
    
ε 3   S33 S34 S35 S36  σ3 
    
 =   (4.9)
    
ε 4  
   S44 S45 S46 
 σ4 
 
    
ε  
 5  S55 S56 
 σ5 
 
    
ε6 S66 σ6

61
 Since both the stiffness and compliance matrices need to be symmetrical along

the main diagonal, only 21 out of 36 constants are needed to characterize the

material. A further simplification was made by modelling the material as or-

thotropic.

Feng et al. [186] printed several blocks of concrete using essentially mate-

rial extrusion and performed simple compressive and flexural tests to determine

strength, Poisson’s ratio, and elastic modulus. As previously discussed the re-

sults were that there was a difference in the responses in all three axis, meaning

the blocks were best represented as orthotropic [186]. These have 3 material di-

rections, which are (from strongest to weakest) [37]:

• the forces that hold the particles together in the same track

• the binding force between adjacent tracks in the same layer

• inter-layer forces

The regions that produce these forces are shown in Figure 4.6.

Figure 4.6: Microstructural details of a 3D printed material.

The orthotropic model, which is assumed to be true under linear elastic de-

62
formations, has been successfully used in the literature [187, 188, 185, 189, 190].

Therefore since 3D printed parts are similar to Figure 4.6 the stiffness matrix for

an orthotropic material is simpler than that for fully anisotropic,

    
σ1  C11 C12 C13 0 0 0  ε 1 
    
σ  
 2  C22 C23 0 0 0  ε 
  2
    
σ3   C33 0 0 0  ε 3 
    
 =   (4.10)
    
σ4  
   C44 0 0  ε 4 
 
    
σ  
 5  C55 0   ε 5 
 
    
σ6 C66 ε6

Inverting and introducing E, ν, and G into Equation 4.10 the result is

    
1
ε 1   E1 − νE212 − νE133 0 0 0  σ1 
    
ε  − ν12 1
− νE323 0 0 0   σ2 
 
 2   E1 E2
    
   ν13
ε 3  − E1 − νE232 1
0 0 0  σ3 
 
 = E3   (4.11)
1
    
ε 4   0 0 0 0 0   σ4 
 
   2G23
    
ε   0 1
0 0 0 0   σ5 
 
 5  2G31
    
1
ε6 0 0 0 0 0 2G12 σ6

Here 1, 2, and 3 represent the material directions. E1 is the stiffness in direction

1 so if only tension is applied in direction 1, Equation 4.11 will reduce to Equa-

tion 4.4. Figure 4.7a shows the labelling convention of normal and shear stresses

that has been used as a basis for this research. The material directions have been

defined as shown in Figure 4.7b. 1 and 2 are parallel to the tracks and 3 is perpen-

dicular to the layers. As can be seen in Equation 4.11, there is no shear coupling,

Rodríguez et al. [191] found that it is present but they concluded that it is weak

and negligible therefore it will be ignored.

63
 (a) Labelling of normal stresses, shear stresses, and material directions from ASTM
standard 5379 [29].

(b) Definition of material directions used in this research, where 1 and 2 are aligned to
the tracks and 3 is parallel to the deposition direction.

Figure 4.7

64
4.2 Material properties

Equation 4.11 requires the knowledge of the following material properties:

• E1 , E2 , and E3

• ν21 , ν13 , and ν32 (the assumption is that Equation 4.11 is a symmetrical ma-

trix)

• G23 , G13 , and G12

Although not specifically designed for AM, existing ASTM testing proce-

dures were used to measure the constants and are described in the following sec-

tions. Each test was performed by printing the relevant coupon in several orien-

tations to test the various material directions. The average and sample standard

deviation s of all the samples in a particular orientation were used to calculate

the material properties and their errors.

4.2.1 Young’s modulus E

This parameter is measured using coupons made according to ASTM standard

D638-14, shown in Figure 4.8 [30], using first type IV then type I coupons, whose

detailed dimensions can be found in Table 4.1. The reasons for changing from

type IV to I will be explained later.

Table 4.1: Dimensions of type I and IV coupons [30].

Parameter Type I (mm) Type IV (mm)

W 13 6
L 57 33
WO 19 19
LO 165 115
G 50 25
D 115 65
R 76 14
RO ... 25
T 4 4

65
 Figure 4.8: ASTM tensile coupons used [30].

Stress-strain curves such as the ones shown in Figure 4.15a were produced for

E1 , E2 , and E3 . The elastic modulus for each constant was calculated by taking

the slope of the linear portion of the curve for each sample then averaging the

slopes. The standard deviation was used for the error.

4.2.2 Poisson’s ratio ν

The basic equation for calculating ν is as follows [30],

−dε t dP
ν= (4.12)
dP dε ax
−ε t
= (4.13)
ε ax

where ε t is the transverse strain, ε ax is the axial strain, and P is the load applied.

Like for E, the test procedure is also outlined in ASTM standard D638-14 [30]

and uses the same samples. It involves measuring axial and transverse strain

as described in section 4.3. A plot was made of transverse versus axial strain

and the linear portion identified. A linear regression was then performed on the

linear portion of each sample’s plot and then the average was calculated to give

the final ν.

66
 Table 4.2: Dimensions of shear sample [29].

Parameter Size (mm)

d1 19
d2 3.8
h 4
L 76
r 1.3
w 11.4

4.2.3 Shear modulus G

To calculate G, the following equation from ASTM standard D5379 was used

[29]:
∆τ
Gchord = (4.14)
∆γ

Here Gchord is equivalent to G, ∆τ is the change in shear stress, and ∆γ is the

change in shear strain. Like E and ν, G is equivalent to the slope of the linear

portion of the plot of shear stress against shear strain.

The two options for measuring this constant were ASTM standard D5379

(Iosipescu test) [29] and D4255 (rail shear method) [192]. The Iosipescu test [29]

was selected because it is specifically designed for composite materials, which

materials fabricated via AM are most similar to, and it provides accurate results

[187]. The sample shown in Figure 4.9a, whose dimensions can be found in Ta-

ble 4.2, is placed in the fixtures shown in Figure 4.9b which have been mounted

on the testing machine.

A shear force P is then applied to the sample and stress and strain data is

collected. The shear strain is calculated with the following equation [29]

γi = |ε +45 | + |ε −45 | (4.15)

where γi is the strain at the i-th data point while ε +45 and ε −45 are the normal

strains at the same data point. A plot of shear stress against strain is then made

as shown in Figure 4.10. The linear portion marked “Chord Modulus Region”

67
 (a) Sample used for shear testing [29].

(b) Testing fixtures used to apply shear stress [29].

Figure 4.9

68
is used to obtain the shear modulus. The average of the shear moduli of all the

samples is then calculated.

Figure 4.10: Shear stress against shear strain showing important features from ASTM standard
D5379 [29].

4.3 Methodology for measuring material properties

4.3.1 Sample orientation and naming convention

Figure 4.11b shows the orientations in which samples were printed, which were

developed using Figure 4.11a from ASTM standard D5379 [29]. They are named

according to the constant the samples were used to measure. Some orientations

can be used to measure two constants like E2 is used to measure ν21 . The material

directions have been defined so that 1 is aligned to the +45 track, 2 is aligned to

the -45 track, and 3 is parallel to the layer deposition direction.

A naming convention was devised, an example of which is ±45-E1-01, which

69
(a) Orientation of material planes used to develop coupon orientations
[29].

(b) Diagram of sample orientations naming system. The white axes

shows the printer coordinate system while the blue axes show the
material directions. Although type IV samples are shown here, type I
were also printed in the same orientations.

Figure 4.11

70
refers to the first (01) sample in the E1 orientation (E1) for the set with the ±45

in-fill. Another example is +45-v31-04, which describes the fourth sample (04)

printed in orientation ν13 (v31) for the set with mono-directional in-fill +45.

4.3.2 Procedure

Two Shimadzu Autograph AGS-X Series Universal Testing Machine (UTM)s were

used for testing, one with a 50 kN load cell and the other with a 10 kN load

cell. The maximum values for the load cells are quite high and if the forces gen-

erated during testing are too small, the load cell would be unable to measure

them accurately. The rule of thumb is divide the load cell rating by 500 to obtain

the minimum force necessary for the reading to be considered valid [193]. For
10000
the 10 kN machine that is 500 = 20 N while the minimum force for the 50 kN
50000
UTM is 500 = 100 N. All experiments for measuring the material constants pro-
duced forces much greater than these values so they can be considered accurate.

There were instances during spring testing where not enough force was gener-

ated therefore a slightly different approach was used which will be discussed in

subsection 6.2.6.

A problem was quickly discovered when it came to measuring Poisson’s ratio

and the shear moduli. The UTMs were equipped with cameras to measure strain

and therefore calculate Poisson’s ratio. The normal procedure was to draw two

parallel lines as shown in Figure 4.12 that were orthogonal to the gauge area.

Trapezium X, the program that was used to control the UTMs and take data, was

able to track these lines as well as the edges of the sample in order to calculate ν.

Figure 4.12: Type I tensile coupon ready with black lines for measuring axial strain.

Sometimes the changes in the sample’s width (the transverse strain) were so

small that Trapezium X could not detect any change, especially with the type

71
IV samples because of their smaller size. Due to the layered nature of 3DP, the

in-built strain detector would also get confused because different parts of the

sample shrunk at different times, creating incorrect data. Therefore another more

accurate method was used as explained below.

4.3.3 GOM Correlate

Digital Image Correlation (DIC) is the technique employed to measure strains in

this research, which has also been used by the literature [194, 38]. The specific

software used here was GOM Correlate, a free DIC package [195]. This program

allowed much greater precision than the video extensometers on the UTMs and

even strain gauges (since it allowed the tracking of individual parts of the sam-

ple’s surface) therefore was used for all testing.

Before testing all coupons were sprayed with a stochastic speckle pattern as

shown in Figure 4.13c. Measurements of the cross-sectional area of the gauge

length were entered into the software then coupons were placed in the UTM and

the test was then started.

Stress, force, and displacement data was recorded at 10 Hz. A video of the test

was taken for strain analysis using the cameras mounted on the UTMs, which

was subsequently converted to 10 frames per second to match the rate at which

data was collected. This was necessary because the frame-rate of the raw footage

varied from 22 to 25 frames per second and was not controllable. Conversion

was performed so as to not compromise the resolution of the video, only the

frame-rate was changed.

An external camera with a higher frame-rate and better resolution could have

been used in lieu of the cameras already mounted on the testing machines but

there was the chance of introducing alignment errors because the optical axis of

the camera has to be orthogonal to the surface of the sample. It would also have

also been difficult to match the start of the strain data, which come from GOM,

and the stress data, recorded by the UTM. The mounted cameras were correctly

72
aligned and Trapezium X ensured that the video started at the same time as the

stress measurements, thereby guaranteeing that the stress and strain data were

synchronised.

After conversion the video was analysed in GOM Correlate. First the area for

analysis had to be selected, as shown in Figure 4.13a. GOM would automatically

verify if the speckle pattern quality was satisfactory by displaying a legend with

different colours representing the quality of the pattern and special attention was

taken to ensure that all samples used in this research had a satisfactory pattern.

The facet size and point distance were values that would be adjusted according

to guidelines provided by GOM. The high accuracy computation method was

used for all analyses, which used a bicubic interpolation method. Figure A.1

shows the analysis of a tensile coupon while shear data was collected as shown

in Figure 4.13c.

In the DIC method, first the pixels in each image are mapped relative to their

neighbours. Then by comparing blocks of pixels from one image to the other,

programs like GOM Correlate are able to build up full field 2D and 3D defor-

mation vector fields and strain maps [196]. More specifically, the program min-

imizes the zero-mean normalized sum of squared difference criterion in gray

intensity between images to find correspondences between the subsets in the

spatial domain [197]. Many shape functions are used to track the changes in

the initially square subset and various software techniques have been developed

that allow sub-pixel resolution. GOM is able to track changes on the surface of

the sample that are smaller than the pixels of the video file, thus it is said that it

has sub-pixel accuracy, which was important for the work done in Appendix A.

For DIC to work effectively, the subdivisions need to be random and unique with

a range of contrast and intensity levels, which is simply achieved by using spray

paint [196].

Axes for strain measurement could be created in any orientation in GOM.

For the shear samples, the x and y axes were rotated to match the orientation of

73
the strain gauges described in ASTM standard D5379. Regardless of which type

of sample was being analysed the output of the DIC process was true strain as

defined in Equation 4.3b, an example of which can be seen in Figure 4.13b.

The strain plot shown in Figure 4.14 is the arithmetic mean of the displace-

ments of each subdivision in the selected area calculated at each frame of the

video using the first frame as a reference. Using the mean produces smoother

plots than looking at individual subdivisions for reasons that are discussed in

more detail in Appendix A.

The strain and stress data are then combined and plotted in MATLAB to pro-

duce the necessary graphs for calculating material properties, Figure 4.15a shows

an example for determining E1 . For all E, ν, and G the linear part of the plot for

each sample is selected and an iteratively re-weighted least squares regression is

performed to calculate the slope with a bi-square weighting function to produce

the regression lines seen in Figure 4.15b. The data points on which to perform

the regression were identified by eye. In the last step the slopes of all the sam-

ples are averaged and the sample standard deviation s is used as a measure of

accuracy. For the example shown in Figure 4.15b the average slope is 3751±512

N/mm2 or GPa.

Some strain softening was found to occur after the linear region during tensile

testing of PLA, which is consistent with Letcher et al. [31], as can be seen in

Figure 4.16a. They also observed no strain hardening after plastic deformation

during tensile testing.

Since only one camera was used for DIC, the resulting strain analysis can only

be 2D, meaning that GOM Correlate was unable to detect any out of plane strain.

This is a potential source of error but there is no way of determining it is effects

using this technique. If two cameras were used, 3D DIC could be performed and

so out of plane strains detected and accounted for. Due to the fact that only 2D

DIC can be performed, special care had to be taken to ensure that the samples

were aligned correctly so as to avoid errors in strain measurements caused by

74
 (a) Example of selection of the area for analysis with pattern quality legend on the right.

(b) Simplified view of the analysis of sample ±45-E3-07 in GOM Correlate indicating significant features.

(c) Strain analysis for sample +45-G13-02. The measurement area can be seen in between the two notches.
The x and y axes have been rotated to produce the required shear strain as described in ASTM standard
D5379.

Figure 4.13

75
Figure 4.14: Output from the strain analysis in GOM Correlate of a tensile coupon showing the
average axial (blue) and transverse (black) strains over time.

out of plane strain. Using the cameras that were already mounted on the UTMs

ensured accurate measurements.

4.3.4 Print settings

An original Prusa i3 MK3, shown in Figure 4.17, was the material extrusion

printer used for the PLA samples. The print settings are shown in Table 4.3.

Table 4.3: Print settings

Feature Setting
Nozzle diameter (mm) 0.4
Extruder temperature (◦ C) 210
Bed temperature (◦ C) 65
Layer height (mm) 0.15
Infill (%) 100
Top and bottom solid layers 0
Nout 1

The slicing program used was Simplify3D (❤tt♣s✿✴✴✇✇✇✳s✐♠♣❧✐❢②✸❞✳❝♦♠✴).

To create a smoother surface, most slicing programs will create solid top and bot-

tom layers, which are printed with modified parameters such as different raster

widths [185]. In order to prevent these layers from affecting the mechanical prop-

76
 50

40
Stress (MPa)

30 1
2
3
20
4
5
10 6
7
0 8

0 0.5 1 1.5 2 2.5 3

True strain ·10−2
(a) Complete stress-strain plots for all samples.

15

1
Stress (MPa)

10 2
3
4
5
5 6
7
8
0 E

0 1 2 3 4
True strain ·10−3
(b) Plots of linear portions of stress-strain curves showing average.

Figure 4.15: Plots of data used to calculate E1 .

77
 (a) Stress-strain curves produced by tensile
testing of 3D printed PLA at different raster (b) Raster orientation directions, 0o (top), 90o
orientations [31]. (middle), and 45o (bottom) [31].

Figure 4.16

Figure 4.17: Original Prusa i3 MK3 [32].

78
erties, they were set to 0.

4.4 Results

4.4.1 Reasons for switching from type IV to type I coupons

Type IV coupons were first used because they were smaller than type I, resulting

in less material usage and shorter printing times. In the course of the research the

decision was made to change to type I because it was found that type IV samples

were too weak in the vertical orientations ν13 and E3 , ν32 , resulting in plots such

as those shown in Figure 4.18. When plotted independently against time, the ax-

ial (Figure 4.18a) strain has an overall linear region. The transverse strain, on the

other hand, has a less clear linear region, a can be seen in Figure 4.18b, which is

generally true of all samples. When plotting the transverse strain against the ax-

ial strain (Figure 4.18c) the data becomes very noisy and it is difficult to identify

a clear linear region. The results in subsection 4.4.2 further discuss this fact.

ASTM standard D638 recommends setting the testing speed to either 5 mm/min

or to a speed where breakage occurs at least 30 seconds after the start of the test

[30]. The first samples had a testing speed of 5 mm/min but because the coupons

were very weak, they broke too quickly and few data points were produced so

the testing speed was lowered.

Samples 07 to 09 were tested at a speed of 0.1 mm/min, resulting in plots

that had more data points as can be seen in Figure 4.19a. Taking all samples into

account, ν32 was measured to be 0.510 ± 0.383, where the standard deviation is

close to 75% of the measured value. Such a high standard deviation is unaccept-

able so the switch was made to type I samples. Type I samples took longer to

print and used more material but produced better data, like that shown in Fig-

ure 4.19b. These were tested at a rate of 0.5 mm/min because like the type IV

coupons they broke too quickly when using the recommended rate. Table 4.4

shows the experimentally measured constants for in-fill with raster patterns +45

79
 ·10−3
3
Axial true strain

0
−2 0 2 4 6 8 10 12 14 16 18 20 22
Time (s)
(a) Time versus axial strain.
·10−3
- Transversal true strain

0.5

−2 0 2 4 6 8 10 12 14 16 18 20 22
Time (s)
(b) Time versus transverse strain.
·10−3
- Transversal true strain

0.5

0 0.5 1 1.5 2 2.5 3

Axial true strain ·10−3
(c) Transverse versus axial strain.

Figure 4.18: Plot for measuring Poisson’s ratio using sample ±45-E3-03, a type IV coupon.

80
and ±45 using type I and shear coupons. The latter were measured at at rate of

1mm/min except for the G31 orientation, where the samples were breaking too

fast so the loading rate was slowed to 0.5 mm/min.

4.4.2 Measured constants

Table 4.4 shows the material constants measured using type I and v-notch coupons

printed with PLA. At least 4 coupons were tested for each orientation. Figure 4.20

compares the measured constants for the two in-fills.

Table 4.4: Measured orthotropic constants for +45 and ±45 in-fill using type I and shear
coupons made with PLA.

Material Constant +45 s ±45 s

E1 (MPa) 3111 183 3011 124
E2 (MPa) 3003 43 3326 189
E3 (MPa) 2725 250 2301 149
ν21 0.327 0.013 0.329 0.009
ν31 0.334 0.033 0.236 0.035
ν32 0.356 0.050 0.243 0.036
G12 (MPa) 1159.1 49.1 1125.5 54.9
G13 (MPa) 974.3 62.3 1006.5 67.3
G31 (MPa) 856.7 59.9 891.2 44.6

As can be seen in Figure 4.20a, the trend for the ±45 in-fill is that E3 is lower

than E1 and E2 , which is expected. E2 for the +45 in-fill is within two error

bars from E1 , whereas the expectation would be that, since in the E2 coupon the

strength comes from the bonding of neighbouring rasters, E2 should be signifi-

cantly lower than E1 but that is not the case. On the other hand, E2 and E1 for

±45 are very similar, which is expected.

The Es and νs have small errors in some cases, 1% in the case of +45-E2 while

the largest is about 15% for ±45-ν13 . Compared to He et al., who gave the elastic

modulus value as 2400 MPa for samples with 100% infill [2], the elastic moduli

measured here were higher. The large errors are most likely due to the sensitivity

of PLA to variations in print parameters and conditions [?], and the relatively

poor resolution of the cameras on the UTMs. For both in-fill types, some coupons

81
displayed a strange behaviour where they expanded slightly in the transverse

directions before contracting as normal. A possible explanation is that this is a

form of stabilization before responding to the stress in the expected way. The +45

in-fill produces higher Poisson ratios in the vertical orientations most likely due

to a smaller amount of air gaps as opposed to the ±45 infill; so material in the +45

samples has less empty space to expand into when under stress thereby resulting

in more visible changes compared to the ±45 samples. Tronvoll et al. [198] found

that the effect of voids on tensile strength is very significant, much more so than

the effect on compressive strength. ABS specimens printed with FFF were found

to have a tensile strength that was 68% lower than that of samples made out bulk

material.

The shear moduli have consistently low errors for all samples and the +45 in-

fill is higher in two out of three cases indicating that this in-fill is stronger when

subjected to shear stresses, a result which can be explained with polymer weld

theory [37]. In the +45 infill all material is deposited in the same direction, result-

ing in adjacent layers that are in continuous contact, as shown in Figure 4.21a,

and thereby leading to stronger welds being created. In comparison the roads in

the ±45 in-fill have discontinuous contact regions, resulting in a weaker material

(Figure 4.21b). For this and other reasons which will be discussed in chapters 6

and 7, the +45 in-fill was selected for the springs printed as part of this research.

The values obtained are similar to those from He et al., around 1000 MPa for the

100% infill samples [2].

The results in Table 4.4 are comparable to those of bulk PLA, where the elastic

modulus is 3500 MPa, Poisson’s ratio is 0.36, and the shear modulus is 1287 MPa

[199], with some differences. Firstly, none of the values are as high as those of the

bulk material indicating that this combination of material and printing technique

is detrimental to the mechanical properties. G12 is closest to the bulk material

and G31 is the furthest because in this orientation the shear is applied along the

layers, the weakest bond. The elastic moduli produce a similar result where the

82
vertical orientations are the furthest from the bulk modulus. ν32 for the +45 infill

is the closest to the bulk material’s ν because this is the only orientation where

roads are perpendicular to the camera during testing, meaning that what is being

measured is the resistance of the deposited roads to stress applied along their

length. Since in this orientation there is the greatest amount of uninterrupted

bulk PLA resisting contraction, i.e. no air gaps or tracks in different directions, it

is reasonable that this orientation has the greatest Poisson’s ratio. Conversely the

±45 in-fill suffers because in the vertical samples half the tracks are deposited
perpendicular to the camera and the other half radially, thus offering a much

lower resistance to contraction.

Summary

In this chapter the material properties and material model were discussed. Then

the techniques used for measuring each material constant were explained, along

with the reason for the adoption of DIC and the switch from type IV to type I

samples. The measured constants confirm the pattern found in the literature: the

bonding between layers is the weakest.

83
 ·10−3

1.5
- Transverse strain

1
1 2
3
4
0.5 5
6
7
0
8
9

0 1 2 3 4
Axial strain ·10−3
(a) Transverse versus axial strain for all ±45-E3 samples using type IV coupons.
·10−3
- Transverse strain

1
0.5 2
3
4
5
0 6

0 1 2 3 4 5
Axial strain ·10−3
(b) Transverse versus axial strain for all ±45-E3 samples using type I coupons.

Figure 4.19

84
 3,500 +45 ±45
E (MPa)

3,000

2,500

E1 E2 E3
(a) Elastic modulus.

0.4 +45 ±45

ν

0.3

0.2
ν21 ν31 ν32
(b) Poisson’s ratio.

1,200 +45 ±45

G (MPa)

1,000

800

G12 G13 G31

(c) Shear modulus.

Figure 4.20: Comparison of measured mechanical properties for ±45 and +45 in-fill using PLA.

85
 (a) (b)

Figure 4.21: Comparison of contact regions between layers in +45 and ±45 in-fills.

86
 CHAPTER 5

MULTI-MATERIAL FINITE
ELEMENT SIMULATION OF 3D
PRINTED MATERIALS

As a reminder, the objective of this chapter is to vary the number of perimeters

for samples subjected to tensile loading and it was achieved via:

• simulation of tensile coupons with various numbers of perimeters and the

most commonly used in-fill

• printing and testing of tensile coupons with the same characteristics

• comparison of the simulations and experimental results

5.1 Modelling of 3D printed materials

3D printed objects are composed of the various parts shown in Figure 5.1. First

to be printed is usually the perimeter, tracks that mark the edges for the object

87
in that layer. There can be any number of perimeters although normally there

are two or three. Then the printer will print the in-fill, which is the inside of the

part and is made according to the specifications of the user. The user can set the

printer to print solid, i.e. with 100% in-fill, layers at the top and bottom of the

object, usually two to three layers at both ends. it is important to note that the

layer height and nozzle temperatures do not usually change while printing the

in-fill or the perimeter.

Figure 5.1: Cross-section of sliced object with various parts coloured in. The grey cylinder on the
right represents the nozzle head.

Objects manufactured via FFF have at least two distinct parts, the perime-

ter and in-fill. Due to their different geometries, they will behave differently to

stress. Therefore the objective in this part of the research was to investigate the

effects of Nout on the mechanical properties of tensile coupons and see if FEA

simulations reflect this.

5.2 Experimental procedure

There are three steps in this series of experiments. The first step is to characterize

the properties of both the in-fill and the perimeter, the results of which have

already been presented in subsection 4.4.2. These can be used with Equation

88
4.11 to create the two compliance matrices necessary for the simulations. The

second step was to perform simulations of tensile coupons in Abaqus/CAE with

various Nout values and to calculate Esim and νsim according to the simulations.

The third and final step was to print tensile coupons with the same number of

perimeters as the simulations, measure Eexp and νexp , and compare them to Esim

and νsim . The print settings were those stated in subsection 4.3.4. To measure the

mechanical properties of the perimeter and the in-fill as accurately as possible,

Nout should be 0 when printing samples. Since Simplify3D does not have this

option, Nout was set to 1, the smallest number possible.

5.3 Simulation and experimental verification

As previously mentioned, the FEA software used was Abacus/CAE. Two ma-

terials were created, one for the in-fill and the other for the perimeter using the

results of the experimental characterization. Figure 5.2 shows an example of a

coupon in Abaqus with Nout = 5. The in-fill region was a separate part from the

perimeters, that were mirror images of each other and the inside surfaces of the

perimeter regions were tied to the sides of the in-fill. The end tabs of the coupons

were not simulated. The thickness of the perimeter regions was the width of the

printer’s nozzle, 0.4 mm, which is also the width of the printed tracks.

Figure 5.2: Abaqus assembly showing the in-fill and perimeters.

89
 Figure 5.3 shows the local material directions for the two regions. Since the

printer deposited the in-fill material in only two directions, the in-fill material

directions 1 and 2 were aligned to the global axes as shown. The perimeters, on

the other hand, change their alignment depending on the shape of the object.

Therefore material direction 1 was set to follow the curve of the perimeter, with

direction 2 defined orthogonally to that and direction 3, which was aligned to

the direction of layer deposition. This means that 1 and 2 can change as stress is

applied in the simulation, which is more realistic.

Figure 5.3: Top down of sample to show material directions for the perimeter (red) and the in-fill
(blue). Direction 3 for both is out of the page.

Five simulations were conducted with Nout = 0, 1, 2, 3, 5. The boundary

conditions of the FEA simulations were determined by the experimental setup,

one end of the simulation samples was fixed while the other end moved at 0.5

mm/min. C3D8R elements were used in the mesh with a size of 0.4 mm, which

was chosen because that is the width of the nozzle. The elastic modulus and

Poisson’s ratio was calculated by extracting the stress and strain data for one of

the surface elements in the middle of the gauge area. A simulation was done

with Nout = 0 for comparison.

The last step was to print tensile coupons and measure Eexp and νexp . Sam-

ples with Nout = 1, 2, 3, 5 were printed, Nout = 4 was skipped because most

users do not use more than 3 perimeters so 5 was considered enough. 5 samples

were printed and tested for every Nout . They were tested on the same Shimadzu

UTMs as in section 4.3 using the same speed as for the simulated samples. Stress

90
data was obtained from the machines while strain was measured using GOM

Correlate.

5.4 Results

The results are shown in Table 5.1. Figures 5.4a and 5.4b show how E and ν

change with Nout .

Table 5.1: Elastic moduli and Poisson’s ratios from simulations and experiments.

Simulation Experiments
Nout E (MPa) ν E (MPa) ± ν ±
0 3011 0.3279
1 3013 0.3252 3011 124 0.3378 0.0045
2 3013 0.3265 2900 11 0.3499 0.0307
3 3014 0.3262 2869 25 0.3492 0.0196
5 3013 0.3266 3040 12 0.3405 0.0282

As can be seen in Figure 5.4a, Nout has an effect on the E of printed samples

since it drops from Nout = 1 to Nout = 3 but then goes back up at Nout = 5.

The drop is very small and indicates that adding perimeters slightly weakens

the part. With 5 perimeters the part is slightly stronger than with 1 perimeter

because the perimeters are aligned to the direction of the stress and as explained

in subsection 4.1.3 the strongest bond is that between the molecules in the plastic.

The simulated samples, on the other hand, are not affected by the number

of perimeters. The characteristics of the sample with Nout = 0 are the same as

those of the material, which is to be expected. There is a tiny increase in the

elastic modulus from Nout = 0 to Nout = 1 due to the reinforcement provided by

the perimeter but after that E remains constant. Although there is a difference,

Eexp for Nout = 2 and Nout = 3 are only about 4% and 5% smaller than their

simulation counterparts.

ν on the other hand is not as affected. Figure 5.4b shows that the simulated

Poisson’s ratio changes slightly from Nout = 0 to Nout = 1 but then remains

91
 ·109
3.1
Simulation Experimental

3
E (Pa)

2.9

2.8
0 1 2 3 4 5
Number of outlines
(a) Elastic modulus.
0.4
Simulation Experimental
0.38
0.36
ν

0.34
0.32
0.3
0 1 2 3 4 5
Number of outlines
(b) Poisson’s ratio.

Figure 5.4: Comparison of simulated and printed samples.

92
consistent. Printed samples show some variation in ν but not as great as the

change in E. The most notable difference between the two sets of results is that

the simulations consistently return a Poisson ratio that is slightly smaller than

the experimental one.

As shown in Table 5.2, the data differ by only a few percentage points. This

indicates that even though the number of perimeters does make a slight differ-

ence in the mechanical properties, it is not very significant. Most users print

objects using manufacturer settings, which set Nout = 2 or Nout = 3 in most

cases. Unless the part is very small, meaning the amount of in-fill is comparable

to the amount of perimeter, the conclusion of this experiment is that FEA ten-

sile simulations can be performed using only the properties of the in-fill without

compromising the results.

Table 5.2: Percentage differences between the experimental and simulated elastic moduli and
Poisson ratios
   
Eexp − Esim νexp −νsim
Nout Eexp × 100 νexp × 100
1 0 4
2 -4 7
3 -5 7
5 1 4

Summary

A comparison of Young’s modulus and Poisson’s ratio measured using simula-

tions and printed samples was made in this series of experiments. The number

of outlines was varied to verify if the simulations agreed with the experimental

results and it was found that they do.

93
 CHAPTER 6

EXPLORATION OF 3D PRINTED
SPRINGS

6.1 Springs and AM

The aerospace industry is drawing great benefits from AM and will do so even

more in the near future as has been explained in chapter 2. Chapter 3 has ex-

plained the potential advantages that mechanisms such as springs have for the

space industry, they can operate in many environments and with harsh condi-

tions for a long time without maintenance.

Springs also come in many shapes and sizes and are very adaptable, they can

be modified to meet very specific needs and their only true limitation is manu-

facturing. As a reminder the objective for this section of the research was to com-

pare springs with different in-fills and investigate the plastic region, in which a

permanent deformation remains after a load is removed. This was done by:

• determining the spring wire cross-section that will print best

94
 • printing and testing a variety of helical springs, both compression and ten-

sion

• exploring the differences and similarities between in-fills and types

• performing cyclical loading in the post-deformation linear region

In a broader sense this section is about how AM affects the properties of springs.

This includes changing the cross-section of the spring material, circle vs square,

since 3D printers can do either. Force-displacement plots were produced in order

to measure the springs’ elastic constant k and were used to determine how AM

affects the springs. Springs were printed with ULTEM 9085 and PLA and were

compared. An interesting result was found which will be discussed later.

6.2 Methodology

6.2.1 Spring types

There are many types of springs, classified based on their structure and load

types which are axial, transverse, and torque. As described before they can also

have non-linear responses to loads. The axially loaded helical or coil spring is the

most common type and is the one chosen for this research as explained in sec-

tion 3.1. Loading in tension and compression were both tested in order to inves-

tigate the widest possible range of springs. Compression springs are slightly eas-

ier to print but have limited pitches and displacements whereas tension springs

are more difficult to print. Figure 6.1 shows the important spring parameters or

dimensions that apply to this research.

6.2.2 Spring wire cross-section

An initial study was performed to see which cross-section printed better. Two

tension springs were printed with identical dimensions: d = P = 7 mm, D = 35

mm, and Na = 5. The springs had different cross-sections where one was a

95
Figure 6.1: Spring parameters, Na refers to the active coils of the spring, those are coils that are
not part of either end of the spring.

circle, Figure 6.2a, and the other a square, Figure 6.2b. The circular cross-section

printed slightly teardrop shaped and required many more supports so the square

was chosen instead since it had smoother sides and higher print quality. Springs

with a square wire cross-section have existed for many years [200] but there is

a further reason for selecting it which is related to the derivation of the spring

stiffness equation. In short, having round wire simplifies the derivation of spring

equations but since 3D printed springs are made of layered material, a circular

cross-section would make the derivation more difficult. This will be discussed in

greater detail in chapter 7.

When switching from the spring with the circular cross-section (Figure 6.2a),

which has a mass of 58 grams, to the square cross-section (Figure 6.3a), with a

mass of 74 grams, the difference is only 16 grams. Smaller springs would have

smaller mass increases when switching from a circular to a square cross-section

so this was considered to be suitable even for the space industry, where mass is

limited and closely monitored, and subsection 8.3.4 discusses a method for fur-

96
ther reducing the mass of springs. A potential way to solve the inaccurate circle

printing would be to make the spring much bigger but this was not considered

since it would have required a lot more material and time to print and the mass

increase would have been too great.

(a) Circular cross-section. (b) Square cross-section.

Figure 6.2: Comparing the print quality of circular and square wire cross-sections.

This is already an example of the design freedom that 3D printing can bring to

spring design. Changing the cross-section from circle to square did not require

a major re-working of the production method, just a change in the Computer

Aided Design (CAD). The final printed spring had the exact square cross-section

as wanted.

Another re-design is shown in Figure 6.3a where the tension springs were

originally designed to have printed grips where to attach the clamps of the UTM.

The grips were at the end of an extension arm so that the tension force was ap-

plied at the centre of the helix. The grips were found to be inadequate since ex-

cessive tightening of the jaws could damage or break them. On the other hand,

if the jaws were not tightened enough the spring could slip out. Therefore the

tension springs were re-designed to have holes at the top and the bottom for the

insertion of screws. This allowed the clamps of the UTM to pull the springs with-

out directly making contact with the spring material, as shown in Figure 6.3b.

97
 (b) Spring with screws as grips in the testing
(a) Spring with printed grips.
machine.

Figure 6.3

6.2.3 Printing orientation and in-fill direction

All the springs in this study, both tension and compression, were printed ver-

tically, with the axis around which the coils processed parallel to the layer de-

position. A simple FEA analysis of the springs was performed in an effort to

determine the stress distribution inside the coils. Using the constants measured

in subsection 4.4.2, two FEA simulations were done of a single coil of a tension

spring in Abaqus CAE. The normal material produced by 3D printers is shown

in Figure 6.4a, which, like in chapter 4, is referred to as ±45 in-fill. The other

material orientation tested was where all the tracks printed were in the direction

of the coil, shown in Figure 6.4b and referred to as mono-directional in-fill or

mono. The coils were subjected to a constant upwards velocity of 0.5 mm/min

to simulate the testing that was going to be done later.

When analysing the result of the FEA, the shear stress in Figure 6.5a is evenly

distributed along the length of the coil. Figure 6.5b, on other hand, shows that

there are areas of higher stress concentrations. This indicated that the ±45 in-fill

98
 (b) Mono-directional in-fill.
(a) ±45 in-fill.

Figure 6.4: Material orientations of a single coil for FEA.

might fail before the mono-directional in-fill therefore springs with both in-fills

were printed and tested to verify the results of the FEA simulations.

(a) Mono-directional in-fill. Uneven meshing

produced areas of higher stress which can be (b) ±45 in-fill.
ignored.

Figure 6.5: Stress distributions generated by a constant upwards velocity of 0.5 mm/min.

6.2.4 Spring index and naming convention

An important parameter in the design on springs is c or the spring index

D
c= (6.1)
d

c gives an indication of how difficult or easy it is to stretch or compress a spring.

A spring index of 3 or lower means that the spring is hard to manufacture con-

99
ventionally. Indices higher than 15 produce springs that buckle and tangle easily

so the optimal range is usually about 4 to 12, which was the range chosen for this

study.

Three sets of springs were printed and tested. The first set was composed

of all PLA tension springs, the second all PLA compression springs, and the

third was composed of ULTEM 9085 compression springs. Springs with mono-

directional in-fill are referred to as mono springs, as previously discussed. The

spring dimensions were chosen to be as varied as possible while at the same time

being fast to print. For the tension springs only d and D were varied while P was

kept constant, so as to reduce the number of variables. For the compression

springs all three were instead varied. Figure 6.6 shows an example of a com-

pression spring. As can be seen they have circles at the top and bottom, referred

to as closed ends, which give the testing machines a surface on which to apply

pressure.

Figure 6.6: Example of a PLA compression spring.

Springs are labelled according to their characteristics and what type of in-fill

they have. For example a spring with d = 5 mm, D = 20 mm, and P = 15 mm is

labelled as d5D20P15. Since the tension and compression springs have different

dimensions, it is not necessary to add an extra term to the label to indicate what

type of spring it is. The only exceptions are a few springs that will be discussed

100
later.

6.2.5 Printer settings

For the tension springs a Cubicon Single 3DP-110F was used but with the same

settings as in subsection 4.3.4. The compression springs were instead printed

on the same Prusa i3 MK3 and with the same settings as in subsection 4.3.4.

The ULTEM 9085 springs were printed on a Fortus 450 mc with factory settings,

except for the in-fill.

The equation to calculate the pitch P is based on the angle θ between the

horizontal and a coil is

P = πD tan(θ ) (6.2)

so for D = 20 mm, which is the smallest compression spring, and θ = 50 then

P ≈ 150 mm. This would mean that a spring with that pitch and having Na = 5

would not fit in any of the printers so P was kept low and supports were used.

The mono-directional in-fill was achieved by instructing the slicing programs

to create a number of outer walls that were at least bigger than d2 . This resulted

in the in-fill shown in Figure 6.7a. Figure 6.7b instead shows one layer of the

normal ±45 in-fill for comparison.

6.2.6 Testing equipment and data generated

The springs were tested using the same UTMs as for subsection 4.3.2. As already

mentioned if a load cell with a capacity much greater than the forces generated

during testing is used, unreliable data might be produced. During the course

of testing some of the springs generated very small forces, especially the com-

pression ones, therefore the UTM with a 10 kN load cell was used, which was

the smallest available. Testing was performed at various speeds, usually in the

range of 50 to 80 mm/min and a series of experiments were done to verify that

the change in speed did not affect the results and will be discussed later.

101
 (a) Mono-directional in-fill. (b) ±45 in-fill.

Figure 6.7: Compression springs in the slicing program.

The data generated by the UTMs was different from that in chapter 4. Force

F was plotted against stroke or displacement, which are the same thing i.e. x in

Equation 3.1. The slope of the linear portions of the F-x graphs would give k as

per Hooke’s law. Tests to breakage and cyclical testing were both performed.

Compression springs could not be broken but they were tested to maximum

compression and also cyclically. The cyclical testing was done over two or three

cycles and only for a small displacement, enough to measure the spring con-

stant. The objective was to stay in the elastic region and to not break the samples

so that they could be used again. As will be shown later, some of the more com-

pliant springs showed that they were greatly affected by creep while undergoing

cyclical testing.

6.2.7 Tension spring testing

This section will discuss the testing procedure for the tension springs. The em-

phasis here was the ±45 in-fill but mono springs were still tested.

102
Test to fracture for ±45 springs

The first set of springs tested were tension springs with ±45 in-fill. The dimen-

sions and spring indices can be found in Table 6.1.

Table 6.1: Dimensions of PLA tension springs.

d (mm) D (mm) P (mm) c

25 6.3
4 30 7 7.5
45 11.3
25 5.0
5 30 7 6.0
45 9.0
25 4.2
6 30 7 5.0
45 7.5

As previously mentioned, the pitch was not varied for these springs so that

the effects of d and D could be observed separately. The first loading rate used

was 10 mm/min with 4 springs per spring index and they were stretched to

breaking, thus revealing the full F-x curve. A second linear region after plastic

deformation was observed, a phenomenon that is known and not limited to he-

lical springs but is not typically explored because once a spring has reached the

plastic region, it is considered not functional any more [33, 200, 201]. In this re-

search the plastic region was instead investigated and the results are presented

in section 6.3.1.

Variation of loading rate for ±45 springs

In this test the two least stiff springs from the previous experiments, i.e. the two

that had the lowest k, were tested at various loading rates to verify the effects of

using a load cell with a rating much higher than what it measured. The loading

rate was slowly increased from 10 to 120 mm/min or until the gradient of the

elastic region became relatively constant. The springs were loaded within the

103
 Figure 6.8: Monotonic and cyclic stress-strain curves for SAE 5160 spring steel [33].

linear region from the previous tests so that they could be re-used. This testing

revealed the relationship between loading rate and k.

Cyclic testing of ±45 springs

After breakage, the spring that showed the most stable behaviour was reprinted

to test the plastic region. The springs were cyclically loaded at different points

in the F-x curve in order to observe their behaviour under cyclic loading after

deformation.

Mono springs: test to fracture and cyclic loading

Two more sets of springs were printed with mono-directional in-fill for compar-

ison with ±45 springs. The first set of mono springs were loaded to breakage at

10 mm/min and their ks measured. The second set of mono springs were loaded

cyclically as with the springs in the previous section.

Comparison of tension and compression mono springs

A last set of tension springs was printed and compared to an equivalent set of

compression springs. They were tested at a higher loading rate of 80 mm/min

104
which was used in the testing of the compression springs.

6.2.8 Compression spring testing

Unlike with the tension springs in subsection 6.2.7, P was varied and the mono-

directional in-fill was the emphasis.

±45 in-fill

Only a few were tested in compression since this in-fill was already investigated

in subsection 6.2.7, therefore the compression springs with this in-fill were a sub-

set of those tested in this section and had the dimensions shown in Table 6.2.

Table 6.2: Dimensions of PLA ±45 compression springs.

d (mm) D (mm) P (mm) c

10
20 15 6.7
3
10
40 15 13.3

10
20 15 4
5
10
40 15 8

Only one spring per combination of dimensions was printed and tested since

it was only a surface study. More extensive testing was performed on a partic-

ular combination of parameters as shall be explained later. All the ±45 springs

were tested to failure, i.e. maximum compression where the coils were almost

touching.

Compression mono springs

The dimensions of the compression springs with mono-directional in-fill are shown

in Table 6.3. Once again one spring per combination of parameters was tested.

105
The exception was spring d5D20P15, of which 5 samples were printed for com-

parison with similar springs made with ULTEM 9085.

Table 6.3: Dimensions of PLA mono compression spring.

d (mm) D (mm) P (mm) c

10
20 12.5 6.7
15
10
3 30 12.5 10.0
15
10
40 12.5 13.3
15
10
20 12.5 5.0
15
10
4 30 12.5 7.5
15
10
40 12.5 10.0
15
10
20 12.5 4.0
15
10
5 30 12.5 6.0
15
10
40 12.5 8.0
15

Three springs were tested to maximum compression to see if the plastic re-

gion could be reached. The results showed that the force-displacement plots

remained in the elastic region therefore all subsequent testing was cyclical and

only to measure the spring constant. The cyclical testing consisted of two cycles

of loading and unloading at between 50 and 80 mm/min since preliminary test-

106
ing on the tension springs had demonstrated that in this speed range the stiffness

of the springs did not change. Most testing was done at the higher speed though

to ensure that the force data collected by the UTM was as reliable as possible.

ULTEM 9085 compression mono springs

The final set of springs to be printed were made of ULTEM 9085, whose dimen-

sions can be seen in Table 6.4. They also underwent cyclical testing like the PLA

compression springs and are differentiated by a U at the end so their label is

d5D20P15U.
Table 6.4: Dimensions of compression ULTEM 9085 springs

d (mm) D (mm) P (mm) c

5 20 15 5.0

6.2.9 Analysis of test data

The collected data was analysed in MATLAB. After plotting the force against the

displacement, linear regressions were performed to obtain the spring constants.

The Curve Fitting Toolbox (CFT) and a custom MATLAB script were used to pro-

duce the linear fits. The same robust bi-square regression formula as in chapter 4

was used to produce the fits.

For the springs that were stretched to breakage, the R-squared value was

used to judge the goodness of the fit when performing linear regressions. Sev-

eral trials were done to attain an R-squared value as close as possible to 1, then

the highest one was used. This was done for both the linear and the post-plastic

deformation regions. The compression springs did not need this procedure since

the displacement was within the elastic region, therefore a linear fit was per-

formed by removing the data at the beginning, end, and turnaround points of

the cycles.

107
6.3 Results

6.3.1 Tension springs

±45 tension springs to failure

The springs were stretched at a rate of 10 mm/min until breakage. 4 different

springs were used for each spring dimension combination. Table 6.5 show the

results along with the average and the sample standard deviations s. k1 is the

spring constant of the elastic region and k2 is the constant for the plastic region.

Figure 6.9 shows the F-x plot for all ±45 tension springs that were stretched to

breakage. In the figure springs with the same dimensions have the same colour.

As can be seen some springs broke before they reached the second linear region

so were not considered in the calculation of k2 .

Figure 6.9: F-x plot for all ±45 tension springs. Sets of four springs with the same dimensions
have the same colour.

The second region was shown to exist in all samples, as is expected, although

it was found to not be completely linear. Figure 6.9 shows that at the end of the

plastic region there is a small rise before breakage on several samples, suggesting

that PLA printed with FFF exhibits strain hardening before rupture. Berzal et al.

[34] observed a similar behaviour in their testing, as can be seen in Figure 6.10,

where two out of the three torsion tests showed a slight rise in the moment gen-

108
Table 6.5: Measured spring constants, average and standard deviations for ±45 tension springs
stretched to breakage. All springs have P = 7 and so it has been omitted.

d (mm) D (mm) k1 (N/m) k1 (N/m) s (N/m) k2 (N/m) k2 (N/m) s (N/m)

641.6 132.4
604.8 107
25 655.9 629.35 23.53 110.1 113.88 12.47
615.1 106
317 47.7
387 95.98
4 30 370.8 356 30.26 89.81 82.39 23.31
349.2 96.06
109.5 59.17
114.1 55.63
45 109.6 108.68 5.24 50.87 54.42 3.76
101.5 52
1446
1643 204.9
25 1614 1533.3 110.81 219.3 206.43 12.17
1430 195.1
839.6 141.8
976.2 159.8
5 30 945.1 897.13 74.60 168.8 155.58 11.49
827.6 151.9
274 90.22
284.3 94.5
45 271.7 268.33 17.56 96.58 95.36 4.17
243.3 100.2
3181 259
3388 285.5
25 2837 3076.3 256.14 296.2 282 16.03
2899 287.3
1791
1895 247.5
6 30 1962 1844 104.55 218.6 241.5 20.57
1728 258.4
538.4 135.9
574.7 132.1
45 495.3 528.73 35.68 110.8 131.63 15.39
506.5 147.7

109
erated by cylindrical samples undergoing torsion. On the other hand, no strain

hardening was observed during the shear testing in chapter 4, which indicates

that the spring shape is what causes the strain hardening.

(a) (b) (c)

Figure 6.10: Comparison between torsion tests and FEA simulations in various test pieces [34].
The plots of the torsion tests are grey in colour.

From Table 6.5 it is possible to see that the k2 s were always smaller than the

k1 s since the spring had already passed the point of plastic deformation. The

k2 s also tend to be less varied than the k1 s and roughly in the same order of

magnitude indicating that after plastic deformation the spring shape stops being

the main source of stiffness since they are permanently deformed. The relative

consistency of k2 s compared to k1 s indicate that the stiffness in the plastic region

is more a consequence of material properties rather than the shape.

As indicated by Equation 6.1, increasing D for a given d reduces k1 , which is

expected and in line with what the spring index indicates. On the other hand, the

springs become less brittle as the mean diameter grows so they break at greater

extensions. For example the d6D45 springs broke between 0.4-0.5 m whereas the

d6D30 group broke at about 0.225-0.3 m. The d6D25 springs broke even earlier,

clustered around 0.2 m (ignoring the one that failed prematurely).

Another behaviour observed is that as D for a given d increases the ultimate

strength decreases, which can be explained by the reduction in stiffness of the

springs. When the base area of the tension springs, characterized by D, becomes

bigger or the coils are made thinner, the stiffness decreases, leading to a lower

resistance to elastic deformation and therefore ultimate strength.

Related to the previous point is the fact that specimens with similar indices

110
had different F-x curves, for example springs d5D25 and d6D30 which both had

c = 5. Although c is a linear factor, the results indicate that D has a stronger

influence on the spring’s response to stress since d6D30’s curve is shallower and

longer than that of d5D25. This is due to the fact that when going from d5D25 to

d6D30, d increases by 1 while D by 5, which is much greater.

Table 6.6 shows all the springs ranked according to their k1 . As expected c

and k1 have an inverse relationship. The springs with D = 45 mm tend to be

the weakest while the smaller D = 25 mm are located more towards the top of

the table. As indicated by Equation 6.1, a spring with D = 45 mm would need a

bigger d than 6 mm in order to have a higher k1 .

Table 6.6: ±45 tension springs ranked according to their measured k1 .

Rank d (mm) D (mm) c k1 (N/m)

1 6 25 4.17 3076.25
2 5 25 5 1533.25
3 6 30 5 1844.00
4 5 30 6 897.125
6 4 25 6.25 629.35
5 6 45 7.5 523.26
7 4 30 7.5 356.00
8 5 45 9 268.325
9 4 45 11.35 105.594

The last observation was regarding the consistency of the measurements made.

The trend is that the error for k1 is smaller than or roughly equal to that for k2

and the standard deviation for k1 never becomes greater than 10% of the average

whereas k2 for d6D45 is close to 12%. This is to be expected because the second

linear region occurs after plastic deformation and therefore the material becomes

less consistent than during the linear region. AM, and FFF in particular, has an

intrinsic process variability that is very hard to completely eliminate therefore

having standard deviations in the range of 5-8% of the average can be consid-

ered acceptable. Out of 36 springs, 4 failed during plastic deformation so that

gives a (4/36) × 100 = 11% failure rate. This is most likely due to the fact that

111
PLA and the printer (Cubicon) are relatively low grade, compared to ULTEM

9085 and the Fortus 450mc, and therefore are more likely to produce flaws that

lead to part failure. Conversely, 100% of the springs fractured after the first lin-

ear region and since most spring applications involve operation in this range, the

material and printer performed in a satisfactory way.

Varying the load rate on more compliant ±45 springs

After the first round of testing, the least stiff springs were selected to verify the

effects of loading rate on k1 , which were d4D45 and d5D45. Since they produced

forces much smaller than the load cell of the UTM, it was necessary to verify if

changing the load rate affected the forces they generated.

Two new d4D45 and d5D45 springs were printed and loaded at a rate of 10,

20, 30, 50, 70, 90, and 120 mm/min or until the effect of the loading rate was

verified to be insignificant. Testing was performed in the elastic region only, 0-

0.006 m range, which was derived from the F-x curves in the first round. The

length of the spring was also measured before each test to verify that it had not

been stretched. The results can be seen in Table 6.7 and Table 6.8.
Table 6.7: Results of load rate verification tests for ±45 tension spring d4D45

Loading Rate (mm/min) k1 (N/m) Length of spring before testing (m)

10 (Round 1) 103.3 0.038
10 (Round 2) 105.8 0.04
20 106.1 0.041
30 105.5 0.041
50 106 0.041
70 107.5 0.041
90 107.1 0.041
120 107.4 0.041

During testing there seemed to be some noise at low loading rates in the

form of small, irregular bumps, shown in Figure 6.11. They were found also at

high loading speeds for d4D45 and during testing in chapter 4, which indicates

that they might be caused by instability in the material of the springs themselves,

112
 Table 6.8: Results of load rate verification tests for ±45 tension spring d5D45

Loading Rate (mm/min) k1 (N/m) Length of spring before testing (m)

10 253.1 0.039
20 258.9 0.041
30 259.8 0.041
50 261.8 0.042
70 263.4 0.042
90 (Round 1) 262.5 0.042
90 (Round 2) 263 0.042
120 263.1 0.042
180 265.4 0.042

perhaps due to 3D printing. This phenomenon was not further investigated since

it lays outside the scope and it was not detected in all samples.

Figure 6.11: Noise present in the testing.

Even with the noise bumps, the overall straightness of the lines was not af-

fected. Tables 6.7 and 6.8 show that although increasing the loading rate did have

an effect on k1 , it was very small. The increase from the smallest to the biggest

k1 for d4D45 was about 3.6% of the maximum k1 and about 4.6% for d5D45. It

is therefore reasonable to conclude that low testing speeds did not greatly affect

the results from section 6.3.1 and they can be regarded as accurate.

113
Cyclic testing in post-deformation region for ±45 springs

In this part of the research two springs with dimensions d5D25 underwent cyclic

testing in the plastic region to increase the understanding of the post-deformation

behaviour of 3D printed springs.

For the first spring, which will be referred to as d5D25-k2-1, the second linear

region was estimated to lie within the range of 0.1-0.14 m according to the values

attained in section 6.3.1. The spring was therefore first stretched to 0.14 m then 8

cycles between 0.14 m and 0.1 m were carried out at a consistent loading rate of

10 mm/min.

d5D25-k2-2 also underwent 8 cycles but in a smaller range: 0.125-0.115 m,

which was close to the midpoint of the second linear region. The results of the

two tests can be seen in Figure 6.12.

Figure 6.12: Results of the cyclical testing in the pseudo-linear region.

The slope between two consecutive cyclical end points is roughly equal to

k1 , which suggests that PLA behaves like an elastic, perfectly plastic material

[202]. During loading and unloading, a hysteresis loop was formed for both

spring specimens where, as can be seen from the decreased distance between

successive hysteresis loops, the deformation due to cyclic loading reduces after

multiple cycles due to work hardening [203]. This occurred quicker with the

114
shorter stroke than with the longer because a smaller displacement results in a

small change in the material and faster work hardening.

Yet deformation has not completely eliminated the effects of the spring’s

structure. The fact that the hysteresis loops stabilize and that their slopes are

similar to k1 indicates that these springs can still function roughly like they are

supposed to even after plastic deformation. This could be used as a safety fea-

ture so that if the springs are stretched too far, assuming they have not broken,

they will roughly behave as intended until replaced. Further testing is needed,

though, to characterize the load bearing behaviour of 3D printed springs with

±45 in-fill after plastic deformation.

Test to breakage of mono-directional springs

Using spring d5D25 as a basis for comparison, the effect of changing the in-fill

was investigated in this experiment. As explained in subsection 6.2.5, the springs

had the dimensions d5D25 but were printed with several outer walls so that they

substituted the normal in-fill. Figure 6.13 shows the result of printing and testing

4 d5D25 mono springs and the measured ks can be found in Table 6.9.

Figure 6.13: Force-displacement curves of d5D25 mono spring to breakage.

Since the mono springs have a longer section after deformation, k2 was mea-

115
 Table 6.9: Measured k1 s and k2 s for d5D25 mono spring to breakage.

Test k1 (N/m) k2 (N/m)

1 1567 288.1
2 1567 279.1
3 1498 271.2
4 1503 271.4
k 1533.75 277.45
s 38.44802 7.996041

sured right after the deformation zone. This was done in order to be consistent

with the springs with the ±45 in-fill. Figure 6.14 is a screenshot of MATLAB’s

CFT showing an example of the area where k2 was measured for one of the sam-

ples. Further analysis of these results will be done in section 6.3.1.

Figure 6.14: Screenshot of the CFT in MATLAB to show where k2 was measured. The red
portions of the plot are ignored by the straight line fitting.

Cyclical testing of mono springs in the post-deformation region

Two d5D25 mono spring were printed and tested in this experiment. From Fig-

ure 6.13 the plastic region was estimated to lie within the range of 0.08-0.12 m.

For the first round of cyclic testing the spring was first extended to 0.12 m then

8 cycles between 0.12 m and 0.08 m were carried out. The second spring was

placed under 8 cycles of loading between 0.095 m and 0.105 m. Both tests were

116
conducted at a consistent loading rate of 10mm/min. The resulting F-x plots can

be seen in Figure 6.15.

Figure 6.15: Cyclical testing of d5D25 tension mono spring. Small displacement: 0.12 - 0.08 m,
large displacement: 0.095 - 0.105 m.

The results obtained for the mono springs closely resembled the trends for the

±45 springs. The hysteresis loops were present and the changes in deformation
became increasingly stable as more cycles were completed due to strain harden-

ing. Therefore, the conclusion can be made that infill orientation does not play

a significant role in determining the post-deformation properties of 3D printed

PLA.

Comparison of ±45 springs and mono springs stretched to failure

The average k1 of the mono springs, see Table 6.9, is almost exactly the same

when compared to the ±45 d5D25 springs from Table 6.5, therefore it would

seem that the spring constant is mainly affected by the inter-layer bonding and

not the in-fill, otherwise there would have been a more pronounced difference.

The k2 s are also relatively close, 277 against 206, but that is to be expected since

deformation can have unpredictable effects.

The standard deviation for both k1 and k2 is lower for the mono spring, 38 and

≈8 respectively, than for the ±45 springs, 110 and 12. Since both types of springs

117
were loaded in the same way, it would seem that having all deposited material

aligned in the same direction leads to a more consistent behaviour, which is cor-

roborated by Figure 6.16. As can be seen there, the mono springs follow the same

curve more closely and for longer than the ±45 springs.

Figure 6.16: F-x plot of all d5D25 springs tested to fracture.

As discussed in subsection 4.4.2, polymer weld theory [37] and the amount of

voids present in the springs helps to explain why the mono spring had a higher

ultimate yield strength than the ±45 ones. Since the mono springs have in-fill

that is aligned to the direction of the helix, voids are going to be created between

concentric rings of material. When stacked one on top of the other the tracks of

deposited filament lay almost exactly one on top of the other, like in Figure 6.7a,

thus the bonding between roads is continuous and such that the inter-layer bond-

ing is mainly dependent on the bond between roads since each road is bonded

to at least 4 neighbours.

The ±45 in-fill instead does not promote a continuous bonding and intro-

duces new voids in the slicing process, like the ones in Figure 6.17, because circu-

lar shapes are not always compatible with the raster pattern. The extra voids are

places where the rasters do not have neighbours so there is no bonding. Overall

the criss-crossing pattern of the rasters makes the ±45 in-fill more susceptible to

118
shearing stresses, which are the main types in springs and this is why the mono

springs were more resistant to failure.

Figure 6.17: Example of ±45 in-fill resulting in a greater number of voids.

The ±45 springs failed at an ultimate tensile strength of 80-100N which corre-

sponded to a stroke of approximately 0.23-0.27m. The mono spring, on the other

hand, fractured and broke after almost twice the amount of load was applied,

180-240N, and at a greater displacement, 0.3-0.35m. Figure 6.18 shows the stark

difference between the two in-fills.

The way the springs broke is indicative of their stiffness. The ±45 springs

typically broke between layers, seen in Figure 6.19a, while the mono ones broke

almost perpendicularly to the filament orientation, shown in Figure 6.19b. This

further proves that the mono springs had stronger inter-layer bonding and so

were able to withstand greater loads.

The mono in-fill pattern is very regular and the cross-section of a mono spring

at any point is also fairly regular, except for the points where one layer ends with

a transverse shell printed perpendicular to the direction of filament. In chapter 7

the assumption is that the direction of the filaments are uniform throughout the

119
 Figure 6.18: Breakage comparison between a ±45 spring (left) and a mono spring (right).

entire cross-section and across all points of the active coils. On the other hand

the ±45 in-fill produces a range of cross-sections depending on where it is taken

along the helix. Both of these features will be further discussed in chapter 7.

Comparison between tension and compression mono springs

In this experiment a comparison was made between tension and a compression

springs with the same dimensions: d5D20P15, with a spring constant of 4. Four

tension and five compression springs were printed, an example of a compression

spring is shown in Figure 6.20. The tension springs were printed using the Cubi-

con printer while the compression springs were printed on the Prusa. Identical

settings were used for both printers to ensure consistency.

The tension springs were tested to fracture while the compression springs

underwent 4 cycles with a constant speed of 80 mm/min. The spring constants

measured for both types can be found in Table 6.10.

As can be seen, the two are very similar. The tension springs have a slightly

lower k but they are within a standard deviation of the compression springs. The

discrepancy is likely due to slight differences in the printers or caused by small

imperfections during the printing process or while removing the supports. This

120
(a) ±45 spring, the indicated lighter areas are fractures along the layers.

(b) Mono spring, where the fracture plane is

perpendicular to the coil.

Figure 6.19: Fractured ends of springs.

121
Figure 6.20: One of the d5D20P15 compression mono springs tested to compare to the tension
springs with the same dimensions.

Table 6.10: Comparison of tension and compression d5D20P15 springs.

Tension Compression
Test k (N/m) k (N/m)
1 2779 2785.85
2 2753 2777.04
3 2743 2810.76
4 2705 2774.73
5 - 2745.64
k 2745 2779
s 31 23

experiment confirms for 3D printed springs what is already known for conven-

tional helical springs: that within the linear region the behaviour of the springs

is going to be the same whether in compression or tension.

6.3.2 Compression springs

Effects of in-fill on compression springs

As previously mentioned, when testing compression springs all 3 dimensions

were varied: d, D, and P. A single spring with each dimension and in-fill com-

bination, showed in Table 6.11, was printed on the Prusa. A single spring per

dimension combination was printed for two reasons. First because the main ob-

jective in this study was to obtain an estimate of various spring constants to com-

122
pare to the spring constants calculated in chapter 7 and not to study the springs’

behaviour in depth. Secondly because the low standard deviations in Table 6.5

showed that one spring per dimension combination was enough.

Spring d5D20P15 was an exception and 5 samples were printed and tested,

as has been discussed. The higher accuracy was necessary because this was the

spring that was compared to the ULTEM 9085 springs and it was also used as a

case study in section 6.3.2.

The ±45 springs were tested to maximum compression, i.e. where adjacent

coils touched. Since none of the the springs left the linear region before reach-

ing maximum displacement, it was then deemed unnecessary to compress the

springs all the way, so the mono springs all underwent 4 cycles with a displace-

ment of 1 cm instead. Table 6.11 shows the results of testing and the difference

between the two in-fills.

Table 6.11: ks for compression springs with ±45 and mono in-fills.

Spring dimensions k (N/m)

Spring d (mm) D (mm) P (mm) c ±45 Mono
d3D20P10 10 318 363
d3D20P15 20 15 6.7 286 320
3
d3D40P10 10 44 41
d3D40P15 40 15 13.3 38 42
d5D20P10 10 2813 2779
d5D20P15 20 15 4.0 2359 2746
5
d5D40P10 10 384 343
d5D40P15 40 15 8.0 367 338

As with the tension springs, the ks are very similar confirming that the in-

fill does not play a significant role in determining the spring constant. The

d5D20P15 spring stands out because there is a difference of about 400N/m be-

tween the two in-fills. This is likely due to an imperfection in the printing process

that caused a weakness in the ±45 spring but is of little consequence when con-

sidering the rest of the springs.

123
Variation of mono compression spring parameters

In this experiment the d, D, and P were each given three levels instead of two

like in Table 6.11 to broaden the investigation. The objective was to verify if

the mono in-fill produced any unusual results. A single spring was printed per

dimension combination and the results from section 6.3.2 were re-used for this

series of testing too. The mono springs underwent 4 cycles with a displacement

of 1 cm from rest and the results can be found in Table 6.12.

Table 6.12: Measured stiffness for mono compression springs.

Spring d (mm) D (mm) P (mm) c k

d3D20P10 10 363
d3D20P12_5 20 12.5 6.7 346
d3D20P15 15 320
d3D30P10 10 108
d3D30P12_5 3 30 12.5 10.0 109
d3D30P15 15 108
d3D40P10 10 41
d3D40P12_5 40 12.5 13.3 46
d3D40P15 15 42
d4D20P10 10 1122
d4D20P12_5 20 12.5 5.0 1112
d4D20P15 15 1142
d4D30P10 10 350
d4D30P12_5 4 30 12.5 7.5 344
d4D30P15 15 353
d4D40P10 10 143
d4D40P12_5 40 12.5 10.0 154
d4D40P15 15 149
d5D20P10 10 2779
d5D20P12_5 20 12.5 4.0 2985
d5D20P15 15 2746
d5D30P10 10 892
d5D30P12_5 5 30 12.5 6.0 873
d5D30P15 15 870
d5D40P10 10 343
d5D40P12_5 40 12.5 8.0 371
d5D40P15 15 338

124
 The findings are not surprising given that they are similar to those for the

tension springs. Like before higher indices gave lower spring constants and vice

versa. Also, two springs with the same spring index have different ks, d3D30

series and d4D40 series. Finally, the variation in pitch does not appear to have a

significant effect but this is likely because the change is small.

The most interesting result from this experiment can be seen in Figure 6.21.

The F-x plots of two springs with low c are shown and the most obvious feature

is the large amount of viscoelastic creep. These two are just an example and other

weak springs displayed this behaviour. As the stiffness of the springs increased,

the hysteresis became less evident.

Figure 6.21: Viscoelastic creep during cyclical testing of two weak mono PLA compression
springs.

Comparison of ULTEM 9085 and PLA springs

This experiment was composed of two series of tests. In the first three d5D20P15

springs were printed using ULTEM 9085 and compared to 5 springs with the

same dimensions fabricated with PLA. In the second 4 types of springs were

printed and compared, one with ULTEM 9085 and one with PLA for every pa-

rameter combination. A Fortus 450mc was used for printing the ULTEM springs

so, as discussed in subsection 3.3.1, the layer height (0.254mm) and other print

125
settings such as nozzle temperature and deposition speed were fixed. The in-fill

type could be changed and so the ULTEM springs (reminder: identified with a

U in their name) were printed using the mono in-fill.

The two sets of d5D20P15 springs were compressed by 1 cm from rest over 2

cycles with a displacement rate of 80 mm/min. Table 6.13 shows the results of

the testing and, as can be seen, the PLA springs have a greater spring constant

than the ULTEM ones, which is an unexpected. Figure 6.22 shows the F-x plots

for all the springs, with the creep removed, and as can be seen the PLA springs

have a steeper slope.

Table 6.13: Spring constants measured from cyclical testing of 5 d5D20P15 and 3 d5D20P15U
springs.

Spring Sample k s
1 2786 0.005
2 2777 0.007
d5D20P15 3 2811 0.005
4 2775 0.006
5 2746 0.003
1 1906 0.005
d5D20P15U 2 1943 0.005
3 1937 0.005

Figure 6.22: Plot of cyclical testing of ULTEM and PLA springs with creep removed.

Further testing was therefore performed on two sets of springs, one printed

126
with ULTEM and the other with PLA, the results of which are shown in Ta-

ble 6.14. The new tests confirmed what had been found previously: the PLA

springs are stiffer than the ULTEM ones. The difference in print settings could

explain this phenomenon but a direct comparison is not possible because PLA

cannot be printed with ULTEM settings and vice-versa. Rodriguez et. al. [?]

found that the tensile strength of PLA and ABS samples fell when increasing the

layer height from 0.1 mm to 0.2 mm, thus it is possible that the different layer

heights, 0.254 vs 0.15 mm, might be one of the factors that weakens the shear

properties of the material. A further experiment was performed to verify this

hypothesis and the results are presented in section 6.3.2.

Table 6.14: Comparison of ULTEM 9085 and PLA springs.

k (N/m)
Spring k PLA − kU (N/m)
PLA ULTEM 9085
d3D40P10 41.0 31.3 9.7
d3D40P15 42.0 33.3 8.8
d5D40P10 343.1 213.5 129.6
d5D40P15 337.9 239.1 98.8

The last comparison made is shown in Figure 6.23a, where the full cyclical

testing of a PLA and ULTEM spring are shown for comparison. Both materials

are equally affected by creep during cyclical testing, therefore ULTEM springs

can be considered overall worse than PLA springs because they are affected by

creep in the same way and are less stiff. Figure 6.23b shows the full testing of all

the ULTEM springs and all are affected by creep.

Effects of layer height on PLA springs

The layer height cannot be changed for ULTEM so the PLA springs were varied

instead. It was not possible to print PLA springs with layers comparable to the

ULTEM ones because above 0.15 mm the quality of PLA becomes inconsistent

with the print settings used in this research, which are designed for 0.15 mm,

127
 (a) Comparison of ULTEM 9085 and PLA springs.

(b) ULTEM 9085 springs.

Figure 6.23: Viscoelastic creep in cyclical testing.

and therefore reliable results cannot be guaranteed. On the other hand, it was

possible to print shorter layers therefore two new d5D20P15 springs were manu-

factured with 0.05 mm and 0.10 mm layer heights and tested like in section 6.3.2.

Table 6.15 compares the stiffness of the two new springs to the ones with 0.15mm

layer height.

As observed by Rodriguez et al. [?] and hypothesised in section 6.3.2, increas-

ing the layer height does seem to weaken the shear properties of the springs,

causing them to become less stiff. The difference is bigger between 0.05 mm

128
Table 6.15: Effects of layer height variation on the stiffness of PLA springs, first round of testing
with one d5D20P15 per layer height.

Layer height (mm) k

0.05 2761.71
0.10 2706.32
0.15 2703.56

and 0.10 mm than between 0.10 mm and 0.15 mm suggesting that layer heights

greater than 0.15 mm would result in lower stiffness losses but the difference is

small and might be due to slight imperfections in the springs. Therefore a new

experiment was carried out where three springs were printed per layer height.

Their dimensions are d7D20P15 with Na = 2, resulting in the stiffest springs

tested in the entire research. This was done in an effort to make the stiffness

differences more pronounced. They were tested for 2 cycles at a speed of 100

mm/min and a displacement of 0.5 cm, the results of which can be found in Ta-

ble 6.16. This smaller displacement was used to avoid deforming the springs.

Table 6.16: Effects of layer height variation on the stiffness of PLA springs; second round of
testing with three d7D20P15 springs per layer height.

Layer height (mm) k (N/m) k (N/m) s (N/m)

25051.20
0.05 26672.60 25804.87 816.70
25690.80
25082.20
0.10 24062.10 25016.60 923.45
25905.50
24983.20
0.15 25373.00 25009.50 351.09
24672.30

As with the previous round of testing, the difference is bigger between 0.05

mm and 0.10 mm than between 0.10 mm and 0.15 mm. The average stiffnesses

for 0.05 mm and 0.10 mm are within one standard deviation of each other but

combined with the results of the d5D20P15 springs it can be posited that increas-

129
ing layer height causes a lowering of the spring stiffness, likely caused by a weak-

ening of the shear modulus of the material. This is unlikely to be the only factor

to create the difference between the PLA and ULTEM springs in Figure 6.22, but

this experiment has shown that layer height does at least play a part and has lead

to the design guideline whereby smaller layers produce stiffer springs.

Summary

In this chapter a thorough investigation of 3D printed helical springs was carried

out. The two in-fills were compared, mono and ±45, and although the difference

in stiffness was not significant, the mono springs ruptured at much greater ex-

tensions than the ±45 springs, which makes this infill the superior choice. Vary-

ing the spring parameters produced stiffnesses that were in line with what the

spring index predicted. Weaker springs were strongly affected by viscoelastic ef-

fects leading to large hysteresis loops during cyclical testing, but the stiffness was

unaffected. Tensions and compression PLA springs were compared and found

to be similar. Stretching into the post-deformation region followed by cyclical

testing revealed that the springs behave semi-elastically even after undergoing

plastic deformation. PLA were found to be stiffer than ULTEM springs and fur-

ther testing revealed that layer height is inversely proportional to stiffness, i.e.

bigger layers produce weaker springs.

130
 CHAPTER 7

DERIVATION OF AN EQUATION
FOR CALCULATING THE
SPRING CONSTANT OF 3D
PRINTED SPRINGS

In this chapter the objective was to derive an equation that calculates the spring

constant of 3D printed springs. This was achieved via the following sub-objectives:

• identify a derivation method

• perform the derivation

• compare the calculated and experimentally measured spring constants

131
7.1 Existing equation for springs

Helical springs with a square cross-section, also called rectangular wire springs,

when manufactured using conventional means cause the cross-section of the

wire to distort during winding due to uneven stresses, as shown in Figure 7.1.

Fabricating a keystone profile can help to promote a rectangular cross-section af-

ter coiling but a better solution would be to use AM since, as shown in chapter 6,

it is very easy to change the cross-section.

Figure 7.1: Change in wire cross-section after coiling for rectangular wire springs [35].

As discussed in subsection 6.2.2, the square cross-section has been chosen for

this research since it results in higher quality prints. The equation that will be

discussed in this section applies to circular springs but is relevant to rectangular

cross-section springs too.

Springs are widely used in many fields and although there are many different

types of springs, the most common one is the helix so that is the one that has

been covered in this research. The existing equations for helical springs have

two assumptions:

1. The material is isotropic

2. The wire’s cross-section is a circle

Isotropic materials are the standard for springs because they are mostly made

with steel that is extruded through a die. There are also polymer springs but

they are relatively rare. The circular cross-section is typical of helical springs but

the rectangular cross-section is usually found in leaf, spiral torsion, and volute

springs.

132
 Springs can also be machined out of metallic tubes using CNC machines to

produce the required helical path and provide the desired elasticity [36]. These

closer resemble 3D printed springs because they can be modified with features

such as threads, clamps and fixing holes, resulting in simpler parts like the ex-

ample shown in Figure 7.2. Although more flexible than traditional springs, ma-

chined springs cannot achieve the shapes that 3D printed springs can like hollow

wire springs, which will be discussed later.

(a) Three piece part

(b) Single part with
with wire wound
machined spring.
spring.

Figure 7.2: Example of simplification of parts using machined springs [36].

7.1.1 Calculating the stiffness of isotropic springs

Elasticity is the ability of a material to return to its original shape after a stress

has been applied, springs are no different and they obey Hooke’s law: F = kx.

As mentioned in section 2.1, it is possible to derive an equation to calculate k

from the spring parameters shown in Figure 6.1,

Gd4
k= (7.1)
8D3 Na

The derivation method for Equation 7.1 will be used as a basis for deriving an

equation for 3D printed springs and will now be explained.

133
Derivation of the shear modulus from the torsion constant for an isotropic

helical spring

The assumption throughout this derivation is that the displacements and rota-

tions are small enough that the spring behaves elastically. When a spring of ra-

dius R is stretched by a force F the wire that the coils are composed of is twisted

along its entire length. There is going to be some stretching along the length of

the wire but that is negligible compared to the twisting.

The spring can be considered to be composed of many identical elements like

the ones in Figure 7.3. When the spring is stretched by an amount dx due to F,

all elements of the spring are twisted by an angle dθ [204].

dθ

dx
dθ

(a) Side view. (b) Front view.

Figure 7.3: Distance that coils move when a spring is stretched.

dx is the length of the arc subtended by dθ so

dx = Rdθ
Z x Z θ
dx = R dθ
0 0

x = Rθ
x
θ= (7.2)
R

The energy stored by a spring can be written in two ways. First, from Fig-

134
ure 3.4 there is the classic expression

1 2
U= kx (7.3)
2

The second expression uses the fact that, as discussed, the wire in the spring

twists and therefore gains torsional energy Ut . There are various versions of this

equation but the one used here is

1 2
Ut = κθ (7.4)
2

Here κ is the torsion constant and is defined as

πr4
κ=G (7.5)
2L

Here G is the shear modulus of the material, r is the radius of the wire, and L is

the total length of the wire that the spring is composed of which can be roughly

calculated using

L = 2πRNa (7.6)

The derivation of Equation 7.5 can be found in Appendix B.

The two expressions for energy can be equated

1 2 1
kx = Ut = κθ 2 (7.7)
2 2

Plugging equations 7.5, 7.2, and 7.6 into Equation 7.7 and solving for k gives an

equation which allows the calculation of the spring constant using the springs’

135
parameters and the material’s shear modulus,

1 2 1
kx = κθ 2 (7.8)
2 2
πr4  x 2
 
kx2 = G
2L R
πr4
k=G
2LR2
πr4
k=G
2 (2πRNa ) R2
Gr4
k= (7.9)
4R3 Na

d D
Substituting r = 2 and R = 2 into Equation 7.9 results in Equation 7.1,

both of which assume an isotropic material and hence why there is only one

shear modulus. The material produced by 3D printing, especially FFF, has been

modelled as orthotropic in this research therefore equations 7.1 and 7.9 are not

useable and ks have to be measured rather than calculated, hence the extensive

testing in this study.

Obviously this is not the most efficient way of creating a springs for applica-

tions discussed in sections 3.1 and 3.2, or in general. It is possible to keep testing

until an appropriate spring is found but having the ability to estimate the stiff-

ness would greatly reduce the design time. Therefore as stated in section 3.3 one

of the objectives of this research was to find an equation similar to Equation 7.1

that predicts the spring constant for a 3D printed spring.

7.2 Derivation method

The two assumptions made about the wire material are:

1. The material is orthotropic

2. The cross-section is rectangular

As previously discussed, modelling 3D printed materials as orthotropic is appro-

136
priate given the high degree of anisotropy that result from the AM process itself,

especially in material extrusion. The cross-section is assumed to be rectangular

for two reasons, one physical and the other mathematical. The physical reason

is that, following the results of the initial samples printed in subsection 6.2.2,

square wire springs were of a higher quality which favoured their adoption in

lieu of the circular ones. Since the springs are made of plastic and given their

small size, increases in mass caused by the square cross-section are minimal and

not considered consequential.

The second reason is that a circular cross-section would not eliminate warp-

ing from the derivation of the stiffness equation like with isotropic materials

[205, 206, 207], see Appendix B. Warping torsion occurs when the cross-section

of a member under torsion is subjected to axial stresses and so displaces axially

along the member [207]. Since the warping function, which describes mathemat-

ically the distortion of the cross-section, has to be found for both cross-sections it

makes no difference which is used. Using a circle would lead to more complex

integrals because the shear moduli, which are assumed to be aligned in the di-

rections shown in Figure 7.5, would have to be defined in polar coordinates or

the boundaries would have to be defined in Cartesian coordinates. Therefore the

square was adopted in the end because the region over which the integration has

to be performed is simpler, described only by four straight lines.

The basic idea in the derivation is to find a form of Ut to plug into Equation 7.7

that is appropriate for the assumptions made above. The derivation begins with

a rectangular prismatic bar of height 2a and width 2b that undergoes torsion

like the diagram shown in Figure 7.4, which is the basis for all the work in this

chapter.

Numerical solutions to torsional problems can be found using methods such

as FEA and this is usually the easiest way, for example using a p-version FEA

[208]. This was not practical for this research, though, since Abaqus/CAE does

not output the torsional constant and anyway the problem was simple enough

137
 Figure 7.4: Rectangular prismatic bar under torsion at one end.

that an analytical solution should have been possible.

As explained in subsection 3.3.3, the mono in-fill was selected for testing since

it is deposited tangentially to the coils. It should be noted that this in-fill is not

perfectly regular, though, there are variations caused by the layering that will be

discussed later. For now the assumption will be that the beam in Figure 7.4 is

homogeneous and looks like Figure 7.5, where the shear moduli correspond to

the +45 values from subsection 4.4.2.

Figure 7.5: Shear moduli for ideal mono in-fill where 1 is the direction tangential to the coils, 3
is parallel to layer deposition, and 2 is orthogonal to the other two.

7.2.1 Warping function vs stress function approach

Generally there are two analysis methods for the type of mathematical problem

in this chapter [209]. The first was introduced by Saint-Venant, which uses dis-

placement components and associated warping functions. This method is gov-

138
erned by the formulation of displacements, from which all other quantities are

derived and the shear strains are represented using the warping function [210].

The second method is by Prandtl and uses stress functions along with the

same displacement formulations that are used for Saint-Venant’s method. The

stresses are formulated using another function, called the stress function, which

allows the elimination of the warping function from the derivation [210].

Both methods were used in this research and require solving of Partial Dif-

ferential Equation (PDE)s of either Laplace or Poisson type. Saint-Venant’s ap-

proach leads to the former while Prandtl’s method yields the latter. Torsion prob-

lems, therefore, are typically a form of Boundary Value Problem (BVP) and can

be solved numerically and sometimes analytically.

If the cross-section of the beam is regular, an analytical solution can usually

be found otherwise numerical solutions are easier. Darilmaz et al. [211] have de-

veloped a method for dealing with arbitrarily shaped orthotropic beams and will

be discussed later. When dealing with rectangular cross-sections Prandtl’s stress

function is usually easier, like in Timoshenko et al. [212]. This is because the

stress function produces Dirichlet (or first-type) boundary conditions, where the

values that a solution needs to take along the boundary of the domain are speci-

fied. Saint-Venant’s method, used in Srinath [213], leads to Neumann (or second-

type) boundary conditions which define the normal derivative of the function at

a boundary. Regardless of which approach is used, the method for solving the

torsion BVP for rectangular sections typically uses Fourier series but an attempt

was made to find a solution without, which is presented in section 7.3.

7.3 Derivation using warping functions

This attempt was made using the method from Slaughter [214] with the starting

equations from Rongqiao et al. [215] because the method in Slaughter derives

the warping function for an isotropic beam, whereas the starting equations from

139
Rongqiao et al. are for an orthotropic beam. According to Saint-Venant’s torsion

theory the displacements are formulated as follows,

u = −θzy

v = θzx

w = θφ( x, y)

where u, v, and w are displacements and θ is the twist angle per unit length.

φ( x, y) is the warping function, which describes the out-of-plane displacements

of the cross-sections of the beam. The strain-displacement relationships in terms

of φ are
   
∂φ ∂φ
γx = θ −y and γy = θ +x (7.10)
∂x ∂y

Using these it is possible to write the following stress-strain relationships

   
∂φ ∂φ
τx = Gx θ −y and τy = Gy θ +x (7.11)
∂x ∂y

There are only two non-zero stress components and they are not dependent on

z therefore the equilibrium equations in the x and y directions are immediately

satisfied. The only equation left is the one that is dependent on the z direction

and therefore the governing equation is obtained,

∂φ( x, y) ∂φ( x, y)
     
∂ ∂
Gx −y + Gy +x =0 (7.12)
∂x ∂x ∂y ∂y

This is a second order PDE. The traction free Boundary Condition (BC)s on the

lateral surfaces require that

∂φ
− y = 0 at x = − a, a (7.13)
∂x
∂φ
+ x = 0 at y = −b, b (7.14)
∂y

140
which are Neumann BCs.

Now that the main equation and the BCs have been established, the solving

process can begin using the method of variation of parameters. A new function

ψ( x, y) is introduced

φ( x, y) = xy − ψ( x, y) (7.15)

which when plugged into Equation 7.12 gives

− Gx ψyy ( x, y) − Gy ψxx ( x, y) = 0 (7.16)

∂2 ψ ∂2 ψ
where ψyy = ∂y2
and ψxx = ∂x2
. Plugging Equation 7.15 into Equations 7.13 and

7.14 gives new BCs

ψx = 0 at x = − a, a (7.17)

2x − ψy = 0 at y = −b, b (7.18)

The next step is to perform separation of variables. The assumption is made

that ψ( x, y) = f ( x ) g(y) so Equation 7.16 becomes

− Gx g(y) f ′′ ( x ) − Gy f ( x ) g′′ (y) = 0 (7.19)

f ′′ ( x ) ′′
This gets rearranged so that Gx f (x)
= − Gy gg((yy)) = λ. The cases where λ > 0
or λ = 0 give trivial solutions where the constants are 0 so φ = 0. When λ < 0

let λ = −k2 where k > 0 so this results in a system of second order differential

equations

k2
f ′′ ( x ) + f (x) = 0
Gx
k2
g′′ (y) − g(y) = 0
Gy

141
 Solving this system leads to

   
kx kx
f ( x ) = C1 cos √ + C2 sin √
Gx Gx
! !
ky ky
g(y) = C3 cosh p + C4 sinh p
Gy Gy

Using the assumption that ψ( x, y) = f ( x ) g(y) and applying the BC 2x − ψy =

0 at y = −b, b gives the following system of equations

! !!
k bk bk
2x = f ( x ) p C4 cosh p − C3 sinh p
Gy Gy Gy
! !!
k bk bk
2x = f ( x ) p C4 cosh p + C3 sinh p
Gy Gy Gy

This is equivalent to saying that f ( x ) g′ (−b) = f ( x ) g′ (b) = 2x so since 2x is odd,

f ( x ) must also be odd. Since cosine is an even function then C1 = 0 so

 
kx
f ( x ) = C2 sin √
Gx

so the system of equations becomes

  ! !!
kx k bk bk
2x = C2 sin √ p C4 cosh p − C3 sinh p
Gx Gy Gy Gy
  ! !!
kx k bk bk
2x = C2 sin √ p C4 cosh p + C3 sinh p
Gx Gy Gy Gy

142
Subtracting these two equations from each other and solving for C3 gives

 
kx
f ( x ) = C2 sin √
Gx
!
ky
g(y) = C4 sinh p
Gy
  !
kx ky
∴ ψ( x, y) = C2 C4 sin √ sinh p
Gx Gy

Now set C2 C4 = A and apply the BC −ψx = 0 at x = − a, a to get

  !
kx ky
ψ( x, y) = A sin √ sinh p (7.20)
Gx Gy
  !
Ak kx ky
−ψx ( x, y) = √ cos √ sinh p
Gx Gx Gy
  !
Ak kx ky
−ψx (± a, y) = √ cos √ sinh p (7.21)
Gx Gx Gy

 
The only non-trivial solution to Equation 7.21 is when cos √ak = 0 so that
Gx
√
π Gx (2n+1)
means that k = 2a therefore Equation 7.20 is a series:

!
∞  
k x k y
ψ( x, y) = ∑ An sin √n sinh pn (7.22)
n =0 Gx Gy

Applying Equation 7.18 again gives

!
∞  
A k bk k x
∑ pn n cosh pn sin √n = 2x
n =0 Gy Gy Gx
∞  
k x
⇒ ∑ Bn sin √n = 2x
n =0 Gx

The orthogonality of terms in the sine series is defined as



Z a 
k x
 
k x
 0,
 if m 6= n
sin √m sin √n dx =
−a Gx Gx 
 a,
 if m = n

143
  
kn x
and since m = n in this case and Bn sin √
Gx
= 2x it is possible to write

a=a
Z a     Z a    
km x kn x km x kn x
sin √ sin √ dx = sin √ sin √ dx
−a Gx Gx −a Gx Gx
Z a      Z a  
kn x kn x kn x
Bn sin √ sin √ dx = [2x ] sin √ dx
−a Gx Gx −a Gx
16a2 (−1)n
Bn a =
(2πn + π )2
16a(−1)n
⇒ Bn = 2 (7.23)
π (2n + 1)2

7.3.1 Final solution for the warping function without Fourier series

Now all the information is present to solve for the warping function. Equa-

tion 7.23 leads to the solution for An ,

√ !
32 Gy (−1)n
p
πb Gx (2n + 1)
An = 3 √ sech p (7.24)
π Gx (2n + 1)3 2 Gy

which is plugged into Equation 7.22 then into Equation 7.15 to give the solution

for the warping function

√ !
∞ 32 Gy (−1)n
p
πb Gx (2n + 1)
φ( x, y) = ∑ 3 √ sech p
n =0 π Gx (2n + 1)3 2 Gy
√ !
π Gx (2n + 1)y
 
1
sin π (2n + 1) x sinh p (7.25)
2 2 Gy

7.3.2 Verification of the solution

The next step is to verify the solution, setting a = b = 0.0025 (the width of the

d5D20P15 springs), Gx = Gy (which simulates an isotropic material) and going

from n = 0 to n = 50 produces the plots in Figure 7.6a. Figure 7.6b shows the

plots of the derivatives of the warping function which give the stresses.

Figure 7.6a shows that the limits from Equations 7.13 and 7.14 are correctly

respected at the corners of the boundary. The derivatives in Figure 7.6b also

144
 (a) Plot of the warping function φ( x, y) without using Fourier series.

(b) Plot of the derivatives of the warping function where orange is φx ( x, y) and blue
is φy ( x, y).

Figure 7.6

145
follow the BCs. A problem occurs when trying to work out the torsional rigidity

GJ, though. Rongqiao et al. [215] give the equation for calculating GJ over a

region R as

ZZ

GJ = Gy (φy ( x, y) + x ) x − Gx (φx ( x, y) − y)y dxdy (7.26)
R

where R in this research is the limits of the rectangle defined by ± a and ±b.

Doing the integral gives a series GJn in terms of R, the shear moduli, and n. For

the solution to be correct the series should be a horizontal asymptote so that

lim GJn = 0 (7.27)

n→∞
n
∴ lim
n→∞
∑ GJn = c (7.28)
1,3,5,...

where c is some value. Plotting the value of GJn for every n in the series produces

Figure 7.7, where GJn does asymptote but not to 0. When plotting the sum of GJn

the sum keeps increasing infinitely, which shows that this solution is not correct

and therefore another approach is necessary.

0.11 2
∑1,3,5,... GJn
GJn

0.1
1
n

9.5 · 10−2
0
0 20 40 0 20 40
n n
(a) Values of GJn for each n. (b) Sum of GJn as n increases.

Figure 7.7

146
7.4 Using Prandtl’s stress function

After the previous attempt, a second method was tried which used Prandtl’s

stress function. As previously explained, this method differs form using warp-

ing functions because instead of solving for displacements the solution is sought

in terms of stresses. It begins the same way as in section 7.3 with Saint-Venant’s

displacement, strain-displacement, and stress-strain relationships. The equilib-

rium relationship is given by the following equation [211]

∂τx ∂τy
+ =0 (7.29)
∂x ∂y

Prandtl’s stress function ϕ( x, y) satisfies Equation 7.29 such that

∂ϕ ∂ϕ
τx = and τy = − (7.30)
∂y ∂x

Combining Equation 7.11 with 7.29 gives the governing PDE to be solved

ϕ xx ϕyy
+ = −2θ (7.31)
Gy Gx

and the traction free BCs are satisfied if

dϕ
=0 (7.32)
dR

or in other words ϕ( x, y) is a constant along the boundary.

The method for finding the solution to ϕ( x, y) was found in Hsieh [216]. The

limits here are different, x goes from 0 to a while y has limits −b/2 to b/2. In this

method the solution is assumed to be of the form

∞  nπ 
ϕ= ∑ Yn (y) sin
a
x (7.33)
n=1,3,5,...

The right hand side of Equation 7.31 can be expanded as a Fourier series between

147
0 and a so that
4θ  πnx 
− 2θ = (cos(πn) − 1) sin (7.34)
πn a

and since cos(πn) = −1 for n = 1, 3, 5, ... Equation 7.31 becomes

ϕyy ∞
ϕ xx 8θ 1  πnx 
Gy
+
Gx
=−
π ∑ n sin
a
(7.35)
n=1,3,5,...

Substituting Equation 7.33 into Equation 7.35 and working out the differentials

gives
1 ′′ π 2 n2 8θ
Y (y) − 2 Y (y) = − (7.36)
Gx a Gy πn

This is a second order Ordinary Differential Equation (ODE) which when solved

for Y (y) gives

√ ! √ !
8a2 Gy θ π Gx ny π Gx ny
Y (y) = + C1 exp + C2 exp − p (7.37)
π 3 n3
p
a Gy a Gy

±b
The constants C1 and C2 are determined from the BCs at y = 2 which for a

simply connected cross-section are ϕ = 0∀( x, y) at the boundary of the region.

Therefore  √ 
πb √Gx n
4a2 Gy θ sech


2a Gy
C1 = C2 = − (7.38)
π 3 n3

Plugging Equation 7.38 into Equation 7.37 then the result into Equation 7.33 re-

sults in the final result for the stress function

∞
√ ! √ ! !
8a2 Gy θ  πnx  πb Gx n π Gx ny
ϕ=− ∑ sin sech cosh −1
π 3 n3
p p
n=1,3,5,...
a 2a Gy a Gy
(7.39)

7.4.1 Verification of the solution

The shear stresses are given by [216]

∂ϕ ∂ϕ
τx = and τy = − (7.40)
∂y ∂x

148
Plotting the stress function and the shear stresses above with a = 0.005, b =

0.0025, Gx = 1159113284, Gy = 974339096, θ = 1, and for n = 1, 3, 5...99 produces

the graphs shown in Figure 7.8.

(a) Plot of the Prandtl stress function.

τzx
τzy

(b) Plot of the shear stresses.

Figure 7.8

The plot of ϕ is zero at the boundaries, as it should be, and the shear stresses

seem reasonable but the real test was confirming if the stress function converges

to a value for the torsional rigidity, which is obtained from the following equation

149
[216],
b
a
Z Z 
2
GJ = 2 ϕ( x, y) dx dy (7.41)
− 2b 0

When combined with the fact that sin2 πn

2 = 1 for n = 1, 3, 5..., Equation 7.41
results in the following expression for the torsional rigidity

∞
√ !!
32a3 Gy θ p q πb Gx n
GJn = ∑ √
5 G n5
πb Gx n − 2a Gy tanh p (7.42)
n=1,3,5... π x 2a Gy

The plots in Figure 7.9 were generated using a = 0.005, b = 0.0025, Gx =

1159113284, and Gy = 974339096. Compared to Figure 7.7, in Figure 7.9a the

values of GJn become smaller as n increases which results in the expected con-

verging behaviour towards an asymptote seen in Figure 7.9b for which the value

is GJ99 = 0.0198752. The convergence behaviour shows that the stress function

gives reasonable answers.

·10−2
·10−2
9.3
Sum of GJn

5
GJn

9.2

9.1
0
0 50 100 0 50 100
n n
(a) Values of GJn for each n. (b) Convergence of GJn as n increases.

Figure 7.9

The next step is to work out the torsional rigidity constant κ which Hsieh

[216] gives as
GJn
κ= (7.43)
ab3

Setting a = 4, b = 2, and Gx = Gy = θ = 1 in Equation 7.43 gives κ = 0.228681

and setting a = b = Gx = Gy = θ = 1 gives κ = 0.140577. These are the same

150
values as those obtained by Hsieh, demonstrating that the stress function works.

7.4.2 Final derivation of the equation for k

There are three equivalent expressions for the energy stored by a beam under

torsion,
1 1 GJn 2 1 Tn2 L
Ut = Tn θ = θ = (7.44)
2 2 L 2 GJn

Here θ is Equation 7.2 while Tn and β are given by

ab3 Gx θβ
Tn =
L
GJn
β= 3
ab Gx

Plugging θ, Tn , and β into Equation 7.44 results in the same thing for all three

equations
GJ n x2
Ut = (7.45)
2LR2
p
then setting this equal to the energy stored by a spring then setting L = Na (πD )2 + P2 ,
which is the length of a helix, and solving for k gives the final equation

GJn
k= p (7.46)
Na R2 (πD )2 + P2

Shear moduli correction

As mentioned before, close analysis shows that the cross-section of the spring’s

wire is not constant, even with the mono-directional in-fill. Figure 7.10a shows

an idealized cross-section if the material was deposited in rectangles, which is

unrealistic but adequate for demonstration. In Figure 7.10b it is possible to see

what the cross-section taken on the right of Figure 7.10c looks like. Notice how at

the bottom of the cross-section there is a continuous track. This is a consequence

of the slicing shown in Figure 6.7a where there are parts of each layer that are

deposited radially to the axis of the spiral. There are in fact radial sections with

151
different lengths so the cross-section will change throughout a single circle. It is

theoretically possible to account for this changing cross-section but it overcom-

plicates the process of finding a numerical solution for the stress function so has

not been considered here. As will be shown later, it was anyway not necessary

to include this correction in Equation 7.46.

(a) Ideal wire cross-section. (b) Wire cross-section taken

from (c).

(c) Side view of a more realistic uncoiled spring.

Figure 7.10

Figure 7.10c shows an approximation of what the spring would look like if it

was uncoiled into a shaft. The layers are not aligned to the z axis of Figure 7.5 and

therefore it is possible that the shear moduli of the material need to be rotated

onto the shear moduli of the shaft in order to give better predictions. Since the

rotation takes place around the x axis in Figure 7.5, G12 = Gx and does not need

152
changing. The angle α in Figure 7.10c is related to P and D

 
P
α = tan−1 (7.47)
2D

and so Gy is

Gy = G31 cos(−α) − G13 sin(−α) (7.48)

Two versions of Equation 7.46 can therefore be written, one that includes the

shear re-alignment so Gy is calculated using Equation 7.48, and the other without,

where Gy = G13 . Both versions were used and compared to the experimental

results in the next section.

A manufacturing solution for the printing of springs that would eliminate the

need for the shear re-alignment is shown in Figure 7.11. Using a robotic manipu-

lator similar to that used in LENS the material could directly be deposited in the

direction of the coil. The two main issues are: first, as can be seen in Figure 7.11,

the deposition head would have to be small enough to fit in between the coils of

the springs; second, the lack of supports could lead to the collapse of deposited

material. Therefore this spring printing technique could be an avenue for future

investigation.

Figure 7.11: Printing of springs using robotic arms.

153
7.4.3 Verification of the final equation for the stiffness of a 3D printed
spring

A comparison of the calculated and experimental spring constants are shown

in Figure 7.12, tables with the values can be found in Appendix C. As can be

seen, the difference between the experimental and calculated spring constants is

small with the biggest being the d3D20P15 spring, which is likely due to some

imperfection during printing. This proves that the derivation method and the

equation work and gives very accurate predictions of the spring constant. In fact

the difference between the experimental and calculated stiffnesses is so small

that it is not possible to determine if the shear realignment is necessary or not

using Figure 7.12.

The average stiffness of the three d5D20P15 springs with a layer height of

0.15 mm from section 6.3.2 was 25,009.50 N/M. Using the equation without the

shear realignment gives a theoretical stiffness of 24,361 N/M, which is about

3% smaller than the experimental value. This result has not been included in

Figure 7.12 because the value is ten times greater than the next stiffest springs,

thus making the bar chart difficult to read, but has been included in Table C.2.

The rightmost columns were added to Tables C.1 and C.2 to see the per-

centage difference between the calculated and measured ks. The formula with

the shear realignment has 7.6% average difference but the standard deviation is

quite large, 5.1%, likely because of the random nature of 3D printing and espe-

cially FFF and PLA. On the other hand without the shear realignment the aver-

age difference is only 4.1% but the standard deviation is bigger than the average,

4.6%. This favours the shear realignment version of Equation 7.46 but there is a

problem. The three levels for the pitch are 10, 12.5, and 15 mm and they seem

to have little effect on the spring constant, whether experimental or calculated.

Upon closer inspection the two versions of Equation 7.46 predict two opposite

behaviours though: with the shear realignment k grows as the pitch is increased

154
 Spring constant (N/m) Spring constant (N/m)

0
1,000
2,000
3,000
0
1,000
2,000
d3D20P10
d3D20P12_5
d3D20P10
d3D20P12_5
3,000
d3D20P15 d3D20P15
d3D30P10 d3D30P10
d3D30P12_5 d3D30P12_5
d3D30P15 d3D30P15
d3D40P10 d3D40P10
d3D40P12_5 d3D40P12_5
d3D40P15 d3D40P15
d4D20P10 d4D20P10
d4D20P12_5 d4D20P12_5
d4D20P15 d4D20P15

155
d4D30P10 d4D30P10
d4D30P12_5 d4D30P12_5
d4D30P15 d4D30P15
d4D40P10 d4D40P10

(a) With the shear realignment.

d4D40P12_5 d4D40P12_5

(b) Without the shear realignment.

d4D40P15 d4D40P15
d5D20P10 d5D20P10
d5D20P12_5 d5D20P12_5
d5D20P15 d5D20P15
d5D30P10 d5D30P10
d5D30P12_5 d5D30P12_5

Figure 7.12: Plots of the measured and calculated spring constants.

d5D30P15 d5D30P15
d5D40P10 d5D40P10
d5D40P12_5 d5D40P12_5
d5D40P15 d5D40P15
Exp k
Exp k

Calc k
Calc k
whereas without the shear realignment the opposite is true. Comparison to the

experimental data is inconclusive because it does not seem to follow a pattern so

will be discussed further in subsection 7.4.4.

The equation was also used on the springs with ±45 in-fill that were tested

in chapter 6. Tables 7.1 and 7.2 compare the experimental stiffnesses to those

calculated using Equation 7.46. As expected both versions of the equation have

a higher average difference than with the mono in-fill indicating that a different

version of the equation would need to be derived for the ±45 in-fill. On the other

hand, the results suggest that the version without the shear correction could be

used to give rough estimates of the shear constant for printed springs with the

±45 in-fill.
Table 7.1: Spring constants for ±45 in-fill with shear correction.

(k calc −k exp )
Spring k exp (N/m) k calc (N/m) k exp × 100
d3D20P10 318 397.841 20.0
d3D20P15 286 406.006 29.5
d3D40P10 44 47.738 8.8
d3D40P15 38 48.863 22.4
d5D20P10 2813 3069.760 8.4
d5D20P15 2359 3132.760 24.7
d5D40P10 384 368.350 -4.2
d5D40P15 367 377.032 2.6
Average 14.0
s 11.8

A comparison has been attempted with the results of He et al. [2], discussed

in section 2.1, but the results were poor because they did not measure the three

shear moduli necessary for use of Equation 7.46. Table 7.3 shows the stiffness

values calculated using Equation 7.46 with no shear correction. G12 and G13 were

estimated to be 800 MPa for 100% infill from Figure 7.13a. Since He et al. did not

include a table of their results, comparing Figures 7.13b and 7.13c is the only way

to verify the calculated values. As can be seen, the calculated spring stiffnesses

seem reasonably close to those measured by He et al., given the estimated shear

156
 Table 7.2: Spring constants ±45 in-fill without shear correction.

(k calc −k exp )
Spring k exp (N/m) k calc (N/m) k exp × 100
d3D20P10 318 379.066 16.0
d3D20P15 286 373.345 23.4
d3D40P10 44 47.8284 9.0
d3D40P15 38 47.6414 20.4
d5D20P10 2813 2924.89 3.8
d5D20P15 2359 2880.75 18.1
d5D40P10 384 369.046 -4.0
d5D40P15 367 367.603 0.1
Average 10.8
s 10.1

moduli. The shape of the plots are also similar, with a steep rise when going from

d = 6.2 mm to d = 7.6 mm. He et al. also varied D, Na , and L, see Figure 2.2b, but

the calculations were not repeated since the exact shear moduli and the measured

values are not known.

Table 7.3: Stiffness calculated using Equation 7.46 with no shear correction and G12 , G13 = 800
MPa.

D (m) P (m) Na d (m) k (N/m)

0.002 14
0.0034 116
0.032 0.01 5 0.0048 462
0.0062 1285
0.0076 2901

7.4.4 ULTEM springs

The last verification was made on a set of springs printed with ULTEM 9085. The

same procedure and samples as for the +45 (mono-directional) infill in subsec-

tion 4.3.2 were used to measure both shear moduli. G13 used the same thick-

ness and testing speed as for PLA, whereas for G12 the samples were made 1

cm thick (see parameter h in Figure 4.9b) and the testing speed was increased to

2 mm/min, the recommended head speed from ASTM standard 5379 [29]. This

157
 3200

Stiffness (N/m)
2400

1600

800

0
2 3.4 4.8 6.2 7.6
(b) Experimental (blue) vs d (mm)
(a) Change in shear modulus calculated (orange) spring
with infill density [2]. stiffness [2]. (c)

was done because when using the previous sample thickness and speed the sam-

ples would produce G12 values that had unacceptably large errors, in the range

of 40-50%. Changing to the new thickness and speed resulted in much more

consistent values, which can be seen in Table 7.4 compared the values for PLA.

As can be seen, ULTEM 9085 in the mono-directional configuration is signifi-

cantly weaker than PLA and has a higher standard deviation, leading to the ob-

servations made in section 6.3.2. Also, G13 for ULTEM is greater than G12 , which

is the opposite of PLA. Overall it would seem that, if not for the fact that ULTEM

9085 is certified for use in space, PLA is the better plastic for springs. Therefore,

an integral part of future investigations into 3D printed springs should be the

measurement of G13 , G12 , and G31 for various materials; so as to create a database

from which the most appropriate material can be chosen.

Table 7.4: Measured shear constants for ULTEM 9085 compared to PLA.

Shear modulus ULTEM 9085 (MPa) ± (MPa) PLA (MPa) ± (MPa)

G13 786.3 127.5 974.3 62.3
G12 574.0 79.6 1159.1 49.1

Several springs were then printed with ULTEM, tested, and their stiffnesses

calculated, a comparison between the experimental and calculated sets of ks is

tabulated in Table 7.5. Included are the d5D20P15 springs used when comparing

ULTEM and PLA springs (section 6.3.2). The version of Equation 7.46 without

shear correction was used, since only G13 and G12 were measured for ULTEM.

There are two conclusions that can be drawn from the results in this table.

158
 Table 7.5: Comparison of calculated and experimental stiffness of ULTEM springs.

(k calc −k exp )
Spring k exp (N/m) k calc (N/m) k exp × 100
d3D20P15 267.4 234.4 -12.4
d3D40P10 31.3 30.0 -4.0
d3D40P15 33.3 29.9 -10.1
d5D40P10 213.5 231.7 8.5
d5D40P15 239.1 230.8 -3.5
d5D20P15 1928.1 1808.3 -6.2
Average -4.6
s 7.3

Firstly, the average error is comparable to those in Tables C.1 (7.6±5.1) and C.2

(4.1±4.6), indicating that the equation provides good estimates not only for PLA.

The second conclusion is that since the stiffness increases with pitch, 31.3 to

33.3 N/m and 213.5 to 239.1 N/m with 10 to 15 mm pitch, the version of Equa-

tion 7.46 with shear correction provides better estimations. Although relatively

few ULTEM springs were manufactured compared to PLA springs, it is possible

to make this conclusion because ULTEM springs have a more consistent print

quality than PLA springs thanks to the higher quality printer and material (For-

tus 450mc and ULTEM 9085) compared to PLA on a Prusa. Therefore the way

to make the most accurate predictions of the stiffness of 3D printed springs in-

volves measuring the 3 shear moduli shown in Figure 7.5: G12 , G13 , and G31 . In

lieu of this, just measuring G12 and G13 will still provide a close estimate of the

stiffness.

Summary

Two equations were derived for calculating the stiffness of a 3D printed spring,

the first uses only G12 and G13 as they are. The second also uses G31 to correct

the shear moduli and align them correctly to the direction of the spring coils.

Comparison with experimental data showed that the equations can predict the

stiffness with about 5-7% error for PLA and 5% for ULTEM 9085.

159
 CHAPTER 8

COSTS, MANUFACTURING
TIME, MATERIAL USAGE, AND
DESIGN FREEDOM

Springs are a well understood component in mechanisms and have been used

for a long time so the following question can be asked: "Why use 3D printing for

springs in lieu of established, reliable methods?". Like for most of the applica-

tions of AM the answer is the design freedom afforded by this novel manufac-

turing technique. For most applications a regular spring is adequate, but there

are many situations where this is not the case and that is where AM may be con-

sidered. The cost, manufacturing time, material usage, and design freedom of

AM applied to springs will now be discussed using the experience gained from

this study.

160
8.1 Material usage

The amount of material used in 3D printing springs is going to be more than that

used for conventional springs, with the exception of machined springs, which

are manufactured from solid pieces of metal. Conventional springs are typically

made via extrusion through a die then winding, resulting in no material wastage,

while AM springs must be printed with supports, resulting in some material

going to waste. In the case of PLA springs, the supports are made of the same

material so the plastic can be recycled into new filament. On the other hand,

printing with ULTEM 9085 always leads to some wasted material because the

supports cannot presently be recycled into new filament.

Table 8.1 shows the material cost of various springs made with ULTEM 9085;

as mentioned in subsection 3.3.1 the canisters of material and support both cost

1320 SGD for 1510 cm3 , resulting in a cost of 0.87 SGD/cm3 . The amount of UL-

TEM 9085 and support material was obtained from Stratasys’ slicing program

for the Fortus 450mc using SMART supports (which minimize the amount of

supports). The cost of PLA springs is not calculated because, an previously men-

tioned, it was only used to develop the model.

Table 8.1: Ratio of support material to ULTEM 9085 used in printing cylindrical springs and
resultant cost. The cost of both ULTEM 9085 and support used for calculations is 0.87
SGD/cm3 .

Spring ULTEM 9085 (cm3 ) Support (cm3 ) Support/ULTEM Cost/spring (SGD)

d3D20P10 12.24 12.48 1.02 21.61
d3D20P15 15.53 16.35 1.05 27.87
d3D40P10 17.52 21.05 1.20 33.72
d3D40P15 20.89 27.41 1.31 42.22
d5D20P10 20.60 17.33 0.84 33.16
d5D20P15 23.43 18.78 0.80 36.90
d5D40P10 33.18 23.61 0.71 49.64
d5D40P15 36.27 31.23 0.86 59.01
Average 0.98
s 0.21

On average the amount of material and support needed is roughly the same,

161
meaning that each spring costs about twice as much as it would without sup-

ports. This can be mitigated somewhat by changing the wire cross-section and

will be discussed in section 8.3. If the material used was instead PLA or ABS, the

cost would of course be much lower, just a few dollars per spring. The cost per

spring does not include labour or machine cost and will be discussed in more

depth in the next section.

8.2 Cost and manufacturing time

The cost and manufacturing time of 3D printed springs are strongly connected

so will be discussed together. Springs are typically not expensive components,

even large ones used in trains are only a small fraction of the total cost [217].

As previously discussed, AM allows greater design freedom so the cost savings

come from the ability to quickly and easily manufacture complex spring shapes,

as well as the fact that there is no need to design and fabricate tools and fixtures

[45].

Using the method from Atzeni et al. [45], an estimate was made of the to-

tal cost to print the biggest (d5D40P15) and smallest (D3D20P10) springs in Ta-

ble 8.2, an exchange rate of 1 EUR = 1.49554 SGD was used to convert the prices.

The higher machine cost per hour, 38 EUR/h, from Atzeni et al. was used so as

to calculate a worst case scenario. The machine operator cost per hour was 14

EUR/h.

The set up and post-processing times were estimated from the experience

gained in this research. ULTEM 9085 support material is not difficult to remove,

since ULTEM is much stronger than the support material, but can be time con-

suming. If several prints of many springs were done, a tool could be developed

and printed to allow quick separation of the springs from the supports. Such a

tool was not developed in this research because most of the printing was done

with PLA and the only way to remove the supports in this case is by breaking

162
them off.

The biggest and smallest springs were chosen so as to calculate the cheapest

and most expensive springs, the others fall between these two. Since D deter-

mines how many springs can fit on one print sheet, a max of 132 can be printed

for all those with D = 20 mm and 42 for those with D = 40 mm. Therefore the

springs considered are: 132 d3D20P10 springs, and 42 d5D40P15 springs. Foun-

dation sheets, which are placed on the platform, come in packets of 20 costing

1100 SGD so 55 SGD each. The cost of the foundation sheet was not added to the

cost of a single spring since they can be re-used depending on the part arrange-

ment and come in packets of 20, meaning that they add a fractional amount to

the total cost.

Table 8.2: Estimated cost of printing springs with ULTEM 9085 using the method from Atzeni
et al. [45].

d3D20P10 d5D40P15
No. of springs per print 1 132 1 42
Machine cost/hour (SGD/h) 56.90 56.90
Build time (h) 2.83 374.43 4.32 181.27
Machine cost (SGD) 161.22 21,305.26 245.62 10,314.07
Machine cost/part (SGD) 161.22 161.40 245.62 245.57
Machine operator cost/hour (SGD/h) 20.96 20.96
Set-up 1 post-processing time per build (h) 1 10 1 6
Machine operator cost (SGD) 20.96 209.6 20.96 125.76
ULTEM 9085 used (cm3 ) 12.24 1616.09 36.27 1523.35
Cost of ULTEM 9085 (SGD) 10.649 1405.998 31.555 1325.315
Support used (cm3 ) 12.48 1646.77 31.23 1311.74
Cost of support material (SGD) 10.858 1432.69 27.17 1141.214
Total material cost (SGD) 21.51 2,838.69 58.73 2,466.53
1 Foundation sheet (SGD) - 55 - 55
Total cost of print (SGD) 203.68 24,408.54 325.30 12,961.36
Total cost/spring (SGD) 203.68 184.91 325.30 308.60

Figure 8.1 shows the normalized cost breakdown from Table 8.2. As can be

seen, roughly four fifths of the springs’ cost is in machine time, caused by long

print times, and printing more springs together results in a lower cost per spring,

which are results similar to those obtained by Atzeni et al. [45]. Similar to Ta-

163
ble 8.1, the ULTEM/support ratio for the maximum amount of springs per print

is about 1 so minimizing the supports would decrease the time and cost. Reduc-

ing the amount of material inside the spring is another way to lower the printing

time, which could be achieved by using a different wire cross-section, as will be

discussed in section 8.3.

100
Machine Machine operator Total material
90 87

79 80
80 76

70
Normalized cost (%)

60

50

40

30

20 18 19

10 11 12
10 6
1 1
0
ing gs ing gs
1 spr 132 sprin 1 spr sprin
P10 P15 5 42
3D20 0P10 5D40 40P1
d d3D2 d d5D

Figure 8.1: Normalised cost breakdown for springs from Table 8.2.

8.3 Leveraging the design freedom of 3D printing to re-

duce the cost per spring

Since supports cause the price per spring to increase, both because of material

cost but mainly due to increased printing time, some considerations have been

made about how to use AM’s design freedom in order to reduce the amount

164
of supports needed per spring. Several methods have been identified and their

applicability will now be discussed.

8.3.1 Deactivation of support generation

In some slicing programs, like Simplify3D, support generation can be deacti-

vated but this typically leads to poorer print quality. In that case the print settings

have to be adjusted to account for this, such as a slower deposition speed and

thinner layers. But this not possible in all slicing programs, for example Strata-

sys’s proprietary slicer for the Fortus 450mc, Insight, does not have an option

to disable supports. Therefore the method of deactivating support generation in

the slicing program is only feasible in specific cases and so not widely applicable.

8.3.2 Changing the spring’s orientation

While leaving support generation active, it is impossible to print springs with-

out supports by changing the orientation, as shown in Figure 8.2. Printing in

orientations other than that used in this research will also change the material

distribution inside the coils, leading Equation 7.46 to become invalid. Therefore

this method for cost reduction is not considered viable.

8.3.3 Increasing the pitch

Another potential method for reducing the amount of supports produced is by

increasing the rise angle, θrise , of the spring coils (the angle between the coils and

the base of the spring), which is found from the arctangent of the pitch divided

by the spring diameter,

P
θrise = (8.1)
D

Three springs with no closed end at the top, which can be printed aside, were

sliced in Simplify3D, shown in Figure 8.3, with support generation set to oc-

cur beyond 45o . As expected, Figure 8.3a causes supports to be generated. The

165
Figure 8.2: Springs in Simplify3D showing that no orientation is possible that does not require
supports. The printer axes are shown for reference.

spring in Figure 8.3b also has supports in equal amount to the spring with θrise =

45o . The final spring, with θrise = 76o (Figure 8.3c), has less supports than the

other two but the pitch is so great that it would make a very stiff spring, therefore

increasing the pitch is not a feasible method to reduce the amount of supports.

8.3.4 Hollow wire springs

The last potential method for lowering the printing time is to manufacture hol-

low wire springs, thus reducing the amount of material needed for printing [66].

A shaft under torsion τ about the central axis z with length ℓ has an angular twist

ϕ given by
ℓτ
ϕ= (8.2)
GJz

where G is the shear modulus and Jz the second moment of area, also known as

the area moment of inertia. Jz for a disk is

πr4
Jz = (8.3)
2

166
 (a) θrise = 45o . (b) θrise = 63o . (c) θrise = 76o .

Figure 8.3: Support generation as θrise is increased.

and, since it is additive,

π (ro4 − ri4 )
Jz = Jzo − Jzi = (8.4)
2

where ro and ri are the outer and inner radii. As can be seen in Figure 8.4, a

hollow tube with r0 =2 mm will have almost the same moment of inertia up to

ri =1 mm. Thus printing a hollow tube would be advantageous in saving both

time and resources.

Vertical round shapes require typically require supports therefore printers

with soluble supports would be advantageous. Otherwise, a solution is to use

a triangular or rhombus cross-section, see Figure 8.5a, the former with a flat

edge parallel to the layers, both of which do not need internal supports if the

sides are steep enough. The rhombus needs less external supports, as shown in

Figure 8.5c, by having a support/material ratio of 0.56 as opposed to 0.95 for

the triangular cross-section, which can be seen in Figure 8.5b. The rhomboidal

wire cross-section springs could also be printed using powder based technolo-

gies such as SLS, assuming a drainage hole is left to let the powder out from

167
 20

Jz
10

0 0.5 1 1.5 2
ri (mm)

Figure 8.4: Second moment of area for a hollow tube with outer radius 2 mm.

inside the spring.

(b) Equilateral triangle wire

(a) The three cross-sections cross-section. ULTEM 9085 (c) Rhombus wire cross-section.
considered: triangle (top), needed: 60.93 cm3 ; support: ULTEM 9085 needed: 55.23
rhombus (middle), square 58.27 cm3 ; support/material cm3 ; support: 31.01 cm3 ;
(bottom). ratio: 0.95. support/material ratio: 0.56.

Figure 8.5: Hollow springs with a wall thickness of 2 mm.

The rhombus can be squashed to a square, like the one at the shown at the

bottom of Figure 8.5a, so that Equation 7.42 can be used. Figure 8.6a shows the

change in torsional rigidity of a spring with a square wire cross-section of outer

diameter 5 mm and with inner diameter di . Similar to Figure 8.4, the torsional

rigidity does not change significantly up to di = 2 mm, therefore a significant

168
amount of material could be removed in order to decrease the cost. Figure 8.6b

shows an example of a compression spring with a hollow, square wire.

0.1

5 · 10−2
GJ n

0
0 2 4
di ·10−3
(a) Change in GJn as thickness of hollow square (b) Compression spring with hollow, square wire
tube decreases. printed out of PLA.

Figure 8.6

The d5D40P15 spring from Table 8.2 has a machine and material cost of 304.35

SGD for a single spring. Table 8.3 shows a comparison of the costs for the same

spring but with a square, hollow wire cross-section and different shell thick-

nesses t shown in Figure 8.5a. Only the machine and material costs are con-

sidered because they have the biggest effect on the cost.

Table 8.3: Comparison of costs for square hollow springs with different shell thicknesses t. The
machine cost per hour is 56.90 SGD and the cost of the ULTEM and support is 0.87 SGD/cm3 .

t (mm) ULTEM 9085 (cm3 ) Support (cm3 ) Build time (h) Machine and material cost (SGD)
Full (2.5) 35.48 33.24 4.05 290.52
2 33.56 32.41 4.15 293.80
1.5 34.32 33.31 4.10 292.41
1 29.19 34.20 3.87 275.43
0.6 22.70 34.86 3.80 266.54

As can be seen, the full spring in Table 8.3 costs less than the one from Ta-

ble 8.2. As t decreases, at first there is an increase in cost but when t is small

169
enough the price drops significantly. Comparing this to Figure 8.6a, when a shell

of thickness 1 mm is used the torsional rigidity has dropped significantly there-

fore a future study could be made to optimize the amount of material removed

while maintaining stiffness. The final thickness greater than 0.5 mm because the

default raster width on the Fortus for ULTEM 9085 is 0.5080 mm.

Summary

In this chapter the costs of manufacturing ULTEM 9085 springs were analysed.

About 80% of the cost comes from machine printing time so four methods for

reducing this were investigated. Of the four, changing the shape of the wire

to a rhombus and printing hollow springs was the only method found to be

universally applicable and that had a tangible impact on the cost.

170
 CHAPTER 9

CONCLUSION

9.1 Summary and Conclusions

AM is a revolutionary manufacturing technology that allows unprecedented de-

sign freedom. It can print a huge variety of materials at high quality. It is finding

a use in an enormous variety of applications, from printing houses to parts for

spacecraft to electronics. Yet it has not yet been adopted universally, there are still

issues that need to be sorted out before it can be fully embraced. The space in-

dustry, amongst others, has recognized the importance of 3D printing and there

is a lot of work underway to ameliorate the issues faced by AM.

Springs are an essential part of mechanisms for storing and releasing mechan-

ical energy. They are easy to manufacture and obtain, finding use in many types

of applications, from the suspensions of vehicles to satellite deployment systems.

By leveraging the design freedom that AM affords, new types of springs could be

developed with shapes and dimensions that were previously hard or impossible

to make. The results of the literature review showed that the application of 3D

printing to springs is a research area that is still very new and only a few studies

171
had been made.

With regards to AMFS a large body of literature is present. AMFS is one

of a few manufacturing areas where a distinction has to be made between the

research that has been done on Earth and in space. The second generation of

space 3D printer is currently on the ISS and up to the present the research done

has been about the 3D printing process and if the parts produced in flight are of

the same quality as those manufactured on the ground, which they are. Some

printing has also been done of tools that incorporate several functions in one,

demonstrating the flexibility of AM.

ISAM is tied to ISM, whose goal is to manufacture and assemble large struc-

tures in space. There are many challenges such as recycling of materials, ability to

move large structures, and being limited to materials that can survive the space

environment. Still there are several proposals for spacecraft that are dedicated to

AM of large metal structures and even one for polymer.

When it comes to AMFS research on the ground, there is a lot of activity.

Many entities, both governmental and private, are performing research in many

different areas. The most prominent one is metal, where the focus is on parts

for engines. Polymer research instead focuses on parts for small satellites like

structures or thrusters. The ability of 3D printers to use almost any material is

being leveraged in ISRU, where research focuses on construction of habitats and

objects for astronauts on other planets or the Moon.

The literature review revealed gaps in AMFS research, mainly the lack of

research into DFAM and a lack of standards specifically for AMFS. Also, as pre-

viously mentioned, there was very little literature on the subject of 3D printed

springs for the space industry. Therefore the objective of this research was to in-

vestigate 3D printed helical springs. A hypothetical, novel CGT design that used

the design freedom of 3D printed springs was proposed as a possible application.

Another potential application of 3D printed springs is their use in deployable he-

lical antennae for CubeSats. This required the modelling of the spring constant

172
of 3D printed springs to avoid the process of printing springs until one was made

that achieved the intended effect.

The research began with the characterization of PLA printed with FFF using

two in-fills, +45 and ±45, according to the orthotropic material model. The DIC

program GOM Correlate was used to measure strain rather than strain gauges,

which allowed a much greater level of detail limited only by the resolution of the

camera used. The measured constants showed that the elastic modulus of both

in-fill was only greatly affected when the loading direction was parallel to layer

deposition and the Poisson ratios had the biggest errors, most likely due to the

fragility of PLA and the difficulty in tracking transversal strain.

After the model was complete an experiment was performed to compare the

simulated and experimental elastic modulus and Poisson’s ratio of samples with

varying numbers of perimeters. The simulations used two different materials

for the perimeter and in-fill, the former being aligned so that the material con-

stants followed the direction of the perimeters. The results showed that although

there was a difference between the experimental and simulated constants, it was

small and not consequential. This experiment showed that in applications where

the size of the in-fill is much greater than the perimeters, simulations of tensile

loading can be considered accurate even if the number of perimeters is varied.

The next step in the research was printing and testing springs. Two sets of he-

lical springs were printed, compression and tension, and their parameters varied.

The in-fill was also varied between the classic ±45 and the mono-directional. The

spring constant of the compression springs was measured with cyclical loading

while in the linear region and revealed that the pitch did not significantly affect

the stiffness of the springs, just like with conventionally manufactured springs.

Comparison between ULTEM 9085 springs and PLA springs, as well as further

experiments, revealed that layer height plays a role in determining the stiffness

3D printed springs with smaller layers producing higher spring constants. The

tension springs could be stretched to breakage therefore cyclical testing was per-

173
formed in the post-deformation region, where the slopes of the F − x graphs

were comparable to the first linear region, suggesting that the spring shape still

had an effect on the stiffness even after plastic deformation. The two different

in-fills were compared and the mono in-fill produced springs that broke much

later than the ±45 in-fill.

Then the equation for the spring constant was developed. The derivation

method involved equating the energy stored by a spring to the energy stored

by a shaft of the same length under torsion. The mono in-fill and a square wire

cross-section were used in the derivation because they simplified the problem

compared to the ±45 in-fill and circular wire. Two approaches were attempted

and the first one, stress functions, did not produce a viable equation. The second

approach instead produced a satisfactory equation that was able to accurately

predict the spring constants of 3D printed springs. A slight variation was made

to the equation to verify if the misalignment of the layers compared to the hypo-

thetical shaft made a difference but comparison to the experimental data did not

reveal any conclusive pattern.

The last part of the research was about leveraging the flexibility of 3D printing

to reduce the cost of manufacturing springs in this way. Only ULTEM 9085 was

considered because PLA was only used to help develop the model. Four meth-

ods were considered and only one was found to be viable in all cases, which was

using a rhombus shaped, hollow wire.

9.2 Contributions

9.2.1 PLA material characteristics

Given that mechanical properties are dependent on the printer and print set-

tings, a contribution is the characterization of 3D printed PLA measured using

two different in-fills: ±45 and +45. Other researchers may still find the informa-

tion useful since all mechanical properties have been measured, including the

174
shear moduli which are not always included in material characterizations. Since

the +45 in-fill has better shear properties than the more traditional in-fill, in ap-

plications where the main loads are shear it would be good to consider this in-fill

type.

9.2.2 Multi-material FEA of 3D printed PLA

The study presented in chapter 5 focuses on comparing the results of FEA and ex-

periments to establish if simulations of samples with varying numbers of perime-

ters give the same elastic moduli and Poisson ratios as real printed coupons when

subjected to tensile loading. The findings were that there is a small discrepancy

that for most applications would not make a difference. Therefore, coupled with

the results from chapter 4, which represent the extreme case of the in-fill being

composed of only perimeters, the main contribution of this investigation is the

following. The E and ν from the simulations did not change with Nout therefore

it is possible to perform simulations of parts with just the regular in-fill and no

perimeters without running into too many issues, being aware that the Poisson’s

ratio from the simulations will be smaller than that of the real part and that the

elastic modulus will vary slightly. This means that users of simulations do not

have to account for varying numbers of perimeters in their FEA models, greatly

simplifying the process especially for complex parts.

9.2.3 3D printing of helical springs

AM enables the fabrication of parts that were previously thought impossible and

the research in chapter 6 has further expanded this possibility. Two major contri-

butions are design guidelines for the printing of springs. Firstly the results show

that the mono in-fill produces a better spring compared to the ±45 in-fill be-

cause, as shown in chapter 4, it has better shear properties. Secondly, the square

cross-section prints better and does not require a keystone profile because there

is no winding. A further contribution was made by the comparison between

175
ULTEM 9085 and PLA springs, which demonstrated that layer height makes a

difference to the spring constant: thinner layers result in stiffer springs. Given

that under cyclic loading the spring still behaves semi-elastically in after plastic

deformation, another contribution is the possibility of leveraging this behaviour

as a safety mechanism if the spring should become deformed, it is still possible

to use the spring but with caution until repairs can be made. The final contri-

bution by this study is the confirmation that, like for springs made of isotropic

materials, 3D printed springs are mainly affected by the wire diameter and coil

diameter and not as much by the pitch, so the spring index c can be applied to

help guide design.

9.2.4 Equation for calculating the spring constants of springs manu-

factured via AM

Building on the work in chapter 6, the successful derivation of Equation 7.46

produces several contributions. First is the fact that, since the orthotropic model

was used as a base, this equation is valid for springs made of any composite

material, not those made via AM. The equation is also quite flexible, it can be

used for rectangular wire cross-section springs by setting a and b to whatever

values are needed. Another contribution, which was previously mentioned, is

that Equation 7.46 greatly cuts down on development times of mechanisms that

use 3D printed springs by allowing the prediction of the spring constant. The

fact that the shear moduli realignment is not needed is also a contribution since

it means that Equation 7.46 can be used as it is without further complications.

Another contribution of this chapter is the fact that Equation 7.46 is a step closer

to the idea of being able to print springs with completely new designs that would

otherwise be impossible manufacture, thus allowing a new generation of mech-

anisms to be developed that take full advantage of AM’s design freedom.

176
9.2.5 3D printed springs and design freedom

The main contribution of this chapter is the determination of the best method for

reducing the cost of 3D printed springs: by using rhombus shaped, hollow wire.

9.3 Future work

Future research could be performed using metal printing, comparing the proper-

ties of printed and conventional springs in an effort to create printing guidelines

that enable high quality printed springs to be manufactured. Also guidelines

could be developed to omit the need for supports or print them so that they are

easy to remove.

9.3.1 Multi-material FEA

Two of three main material constants were covered in this study: E and ν. There-

fore the logical area of expansion for this research would be to perform the same

experiments but for G. This would verify if simulations of shear are also accu-

rate in the same way as it has been done for the elastic modulus and Poisson’s

ratio, creating a complete set of guidelines for designers and engineers with re-

gards to the differences between simulations and experiments when changing

the numbers of perimeters for a printed part.

9.3.2 Investigation of 3D printed springs

An area of further research could be in determining the best method for printing

springs with round wire cross-sections. Another area for future investigation is

high-volume cyclical loading of springs to verify their integrity and performance

when undergoing many cycles of loading and unloading. The final area for fu-

ture investigation is the comparison of PLA and ULTEM 9085 tension springs

stretched to breakage to investigate the post-deformation behaviour of ULTEM.

177
9.3.3 Modelling the spring constant of springs manufactured via AM

The most obvious area for expansion here is determining if the shear correction is

necessary or not. Apart from that, using the same principle of equating energies

it should be possible to derive an entirely new equation that calculates the spring

constant of spiral and torsion springs, since in this case instead of a shaft under

torsion they are beams undergoing bending. The other area for future work is

expanding the equation to take into account variations in the diameter and shape

of the wire as well as changes in the diameter of the coil, for example in the case of

cone springs. For this application the aforementioned method for calculating the

torsion of an arbitrarily shaped orthotropic beam would prove invaluable. The

last area of expansion would be to derive an equation for a spring with circular

cross-section, which has no warping but the anisotropy would still need to be

taken into account.

9.3.4 Design freedom

There is the possibility of printing other types of springs other than helical, like

conical, spiral, or torsion. Making use of 3D printing’s design freedom new

spring designs could be explored, for example ones whose cross-section changes,

either in diameter or shape. Or it would be possible to print single use springs

that lock into place once they reach a predefined length.

178
Appendices

179
 APPENDIX A

INVESTIGATION OF
LOCALISED STRAIN
OSCILLATIONS DURING SLOW
TENSILE LOADING

The content of this appendix has been included to provide a record of a phe-

nomenon that has been observed during testing in chapter 4 and subsequently

investigated. No definite conclusions can be drawn with regards to the phe-

nomenon’s origin or mechanisms, only the observable effect, hence why it has

been placed as an appendix instead of a main chapter. The investigation of

the phenomenon also lies outside the research on 3D printed springs. What is

reported here is what has been observed during several experiments and the

few definite conclusions that can be drawn from the observation of the phe-

nomenon’s effect but, as stated before, no insight was gleamed as to how the

180
phenomenon comes about and how it generates the observable effects.

A.1 Observation of strain oscillations

The procedure for extracting the strain data from the filmed experiments has

been described in subsection 4.3.3 but will be summarised here:

1. During testing, videos are taken of samples

2. The videos are converted from their native frame-rate to 10 frames-per-

second so as to match the rate at which stress data was collected, no reso-

lution was lost

3. The videos are individually imported into GOM Correlate

4. The area of the sample is selected where the strain is to be measured (the

coloured area in Figure A.1)

5. For the shear coupons, a new coordinate system is created to comply with

ASTM standard D5379 [29]

6. The strains in the x and y-directions are computed and exported

7. A linear fitting is performed in MATLAB on the stress-strain data, which

gives the wanted material constant

During the experiments in chapter 4, when extracting the strain data there

are two main methods: deviation labels and average. When using the average,

GOM calculates the strain for every part of the sample’s surface and takes the

average. This is repeated for every frame and when plotted over time the data

results in a line like the one shown in Figure A.2b.

The other method for extracting strain data is via deviation labels. It is possi-

ble to mark any number of specific points on the surface of a sample for tracking,

called deviation labels, such as the one shown in Figure A.2a. GOM is then able

181
Figure A.1: Type IV sample in the DIC program. The red area corresponds to the gauge area and
is where the strain was measured. The green arrow (y-axis) shows the direction of axial strain
while the red one (x-axis) shows the direction of transverse strain.

182
to generate the strains over time for the deviation labels along any axis. As can

be seen in Figure A.2b, when the strain for the point in Figure A.2a is plotted

over time, it does not form a straight line, instead oscillating as the sample is

stretched. For this reason, the average strain was used in section 4.3 instead of

the data from single deviation labels.

The oscillatory behaviour for the single point observed in Figure A.2b is

clearly not noise since the pattern is too regular and the amplitude of the signal is

much bigger than the noise. The point oscillations were found in tensile samples

but not the shear samples, because the data resulted in plots that were too noisy

to be useful, therefore all analysis was carried out on the tensile coupons. Fig-

ure A.3 shows examples of point oscillations obtained from samples in the E1 , E2 ,

and E3 orientations. As can be seen, the signal from the E3 coupon produced a

more consistent strain wave pattern, the peaks are much clearer and well spaced

out, and the amplitude of the oscillations is relatively consistent. The drawback

is that coupons in the E3 orientation break earlier than the others so less data

could be collected per sample. Despite this, more discernible wave properties re-

mained a priority for analysis so this orientation was used regardless. It is likely

that the other orientations also produce the oscillations but it was not possible to

conclusively identify them in this study. Bartolai et al. also used DIC in their re-

search but seem to not have observed this phenomenon since they don’t discuss

it [194].

Another example of the oscillations is shown in Figure A.4, where the axial

and transverse strains for a point on an E3 sample are shown. As can be seen,

the plot of the strain in the x-direction, the transverse strain, has a greater signal-

to-noise ratio so it is difficult to determine if the oscillations are occurring or not.

The oscillations are instead much clearer for the axial strain, which is measured

parallel to the layer deposition direction. Similar plots were observed in all E3

tensile samples, even the type IV ones, although the oscillations were more diffi-

cult to identify. Therefore, all experimental analysis was performed on samples

183
 (a) Example of a deviation label, identified
by the green label, on the surface of sample
±45-E3-07.
·10−2

0.5

0
True strain

−0.5

εy
−1 ε y for deviation label

0 10 20 30 40 50 60 70 80
Time (s)
(b) Average strain in the y direction compared to the strain for the point shown in (a),
also in the y-direction, for sample ±45-E3-07.

Figure A.2

184
 ·10−3
·10−2 ·10−2

True strain
True strain
2

True strain
2
5
1
1
0 0
0
−1
0 100 200 300 0 200 400 0 20 40 60
Time (s) Time (s) Time (s)
(a) E1 . (b) E2 . (c) E3 .

Figure A.3: Examples of point oscillations from tensile samples in different orientations.

printed in the E3 orientation and parallel to the layer deposition direction, which

is also the axis along which the stress was applied. This is the weakest of the ma-

terial directions from Figure 4.7b according to the result in Figure 4.20a, which

is expected. The phenomenon has been labelled localised strain oscillations, or

just strain oscillations, due to the fact that the strain values oscillated over time

as stress was applied.

·10−2 ·10−2
1
1
True strain

True strain

0.5 0.5

0 0

0 20 40 60 80 0 20 40 60 80
Time (s) Time (s)
(a) x (transverse) direction. (b) y (axial) direction.

Figure A.4: Strains for a single point on sample ±45-E3-07, where the axes are aligned like in
Figure 4.13b.

185
A.2 Methodology

A.2.1 Tensile coupon specimens

Three different types of samples were used in total: type I coupons, type IV

coupons, and shorter, thicker coupon based on type I. All samples were printed

with the settings shown in Table 4.3. Table A.1 compares the dimensions of the

type I and redesigned type I coupons.

Table A.1: Dimensions of redesigned type I coupon compared to the regular type I (the
dimensions referenced can be found in Figure 4.8).

Dimensions Type I (mm) Redesigned type I (mm)

W 13 13
L 57 30
WO 19 19
LO 165 120
G 50 24
D 115 88
R 76 76
T 4 10

The orientations initially investigated were E1 , E2 , and E3 from Figure 4.11b.

As shown in Figure A.3, it was found that the E3 orientation produced the clear-

est oscillations therefore that was the main orientation used, the modified type

I coupons were printed in this orientation. The procedure hereafter described

pertains only to coupons the E3 orientation.

Some of the samples were marked with black ink in an effort to emphasize the

layers during DIC as shown in Figure A.5a. These areas were not used for data

gathering as it was feared that the black markings would interfere with GOM

Correlate and produce invalid data. Various test speeds below 1 mm/min were

used and the results will be discussed later.

186
(a) Example of black markings on the right (b) Testing on the UTM.
of the sample to emphasize the layers.

Figure A.5: Modified type I coupon details.

187
A.2.2 Analysis of experimental data

The data gathered during this series of experiments was analysed similarly to

chapter 4. The videos taken by the UTM were first converted to match the rate

of data gathering then imported into GOM Correlate. Within GOM Correlate,

deviation labels were then placed on the area being analysed, these show the

strain at a particular location and can be tracked individually, which is how the

phenomena of strain oscillations was discovered in the first place. Depending on

the type of analysis the labels can be placed anywhere in the analysis area.

One special way that the labels were placed was by referencing the marked

area at the edge of the coupons where the distance between each black line will

correspond to a single layer height or 0.15 mm. With this reference distance set, a

series of points can be created, one per layer in an effort to try to understand how

the strain oscillations behave. Figure A.6 is an example of a series of deviation

labels and their outputs.

Figure A.6: Example of a series of deviation labels aligned to layers (left) and their strain-time
plots (right), where they are all shown together.

One of the limitations faced was that the strain oscillations were not consis-

tent over the whole sample, the clarity of the oscillations varied greatly as can be

188
seen in Figure A.7, which shows the axial strain against time of 5 deviation labels

placed in adjacent layers. Points 1 and 2 have clear oscillations from 45 seconds

on while point 3 doesn’t seem to have any discernible oscillations. Then points

4 and 5 once again have some oscillations from 50 seconds onwards. Locating

points that displayed good oscillations was often a matter of generating tens of

points in a grid pattern and then finding ones that had clear, consistent oscilla-

tions. Moreover there was significant displacement noise in certain regions of

the samples which would hamper analysis. The variation in signal quality of the

points is likely due to the printing process resulting in an inhomogeneous mate-

rial where there are slight differences from layer to layer and track to track. This

is turn would affect the ability of different parts of the sample to react to stress

which, along with interactions with neighbouring parts, result in the inhomoge-

neous oscillations seen.

Something to note is the comparison between the displacement and the strain

for single points. Figure A.8a shows the plot of displacement against time for 5

points on the surface of sample ±45-E3-01. GOM Correlate can output displace-

ments by giving it a gauge distance, which involves entering the distance be-

tween two points on the surface of the sample. In this case two vertically aligned

points on adjacent layers were used so the distance between them was the layer

height, 0.15 mm.

The plots of the displacements are then compared with the plots of the strains

for the same points in Figure A.8b. As can be seen, the two sets of plots are very

different in behaviour. The displacement plots, i.e. the distance that a point

moves from its initial position, create straight lines as is expected, since the head

speed is constant. The strain plots, instead, show the oscillatory behaviour that

has been discussed, although the oscillations are clear only for points 5, 2, and 1.

This discrepancy in the behaviours of the two plots cannot be explained with the

information available here and has been described only for completeness.

Once points had been located with some clear oscillations, the strain-time

189
 ·10−3 Point 1

0 10 20 30 40 50 60 70
·10−2 Point 2
1

0.5

0
0 10 20 30 40 50 60 70
·10−3 Point 3
8
True strain

6
4
2
0
0 10 20 30 40 50 60 70
·10−2 Point 4
1

0.5

0
0 10 20 30 40 50 60 70
·10−2 Point 5

0.5

0
0 10 20 30 40 50 60 70
Time (s)

Figure A.7: Example of a series of 5 deviation labels aligned to layers.

190
 Point 1 ·10−3 Point 1
0.4 5

0.2 0

0
0 20 40 60 0 20 40 60
Point 2 ·10−3 Point 2
0.4
5

0.2
0

0
0 20 40 60 0 20 40 60
Point 3 ·10−3 Point 3
Displacement (mm)

0.4 6
True strain

4
0.2 2
0
0 −2
0 20 40 60 0 20 40 60
Point 4 ·10−3 Point 4
0.4
5
0.2
0
0
0 20 40 60 0 20 40 60
Point 5 ·10−3 Point 5
0.4
5
0.2
0
0
−5
0 20 40 60 0 20 40 60
Time (s) Time (s)
(a) Displacement over time in the axial direction. (b) True strain over time in the axial direction.

Figure A.8: Plots of displacement and strain for 5 deviation labels on the surface of sample
±45-E3-01.

191
data was imported into MATLAB. Here a Savitsky-Golay filter was used to re-

move excessive noise in the readings which can be seen in Figure A.9. Determin-

ing how much noise to remove was done through a trial and error approach by

comparing the cleaned signal to the original but every effort was made to not

lose important information. Next the linear trend caused by the movement of

the jaws during testing was removed as shown by the de-linearized line in Fig-

ure A.9. Following this, the wave properties of the oscillations were analysed in

MATLAB.

·10−2

Original signal
Savitsky-Golay filter
1 After de-linearization
True strain

0.5

0 20 40 60 80
Time (s)

Figure A.9: Example of processing of noisy strain-time data within MATLAB.

MATLAB analysis

This involved measuring the amplitude and approximate period of the oscilla-

tions, which was achieved first with the ❢✐♥❞♣❡❛❦s function and produced plots

like the one in Figure A.10. Finding the approximate period was difficult since

the ❢✐♥❞♣❡❛❦s function occasionally detected secondary peaks due to residual

signal noise even after applying the filter, which are circled in red in Figure A.10.

The false peaks were ignored for all analysis and calculations.

192
Figure A.10: Example of peaks found on a filtered and de-linearised signal, primary (green) and
secondary peaks (highlighted in red) are shown.

MATLAB’s CFT was also used in an attempt to fit a sinusoidal curve to the

oscillations and obtain the wave properties. The coefficient of determination,

also called R2 , was used to gauge how well the fitted wave matched the original

signal. The coefficient of determination is the proportion of the variance in the

dependent variable (true strain) that is predictable from the independent variable

(time). Fits such as the one in Figure A.11 were typical and R2 was never higher

than 0.8 due to the shifting pattern of the signal. The next section will examine

the results obtained in the experiments as well as the subsequent analysis and

modelling in MATLAB.

A.3 Results

A.3.1 Analysis of type I coupons

Visual comparison of oscillations produced by type I samples tested at 1 mm/min

and 0.5 mm/min showed that slower speeds produce better oscillations for anal-

193
 Figure A.11: Fitting a sinusoidal function using the CFT.

ysis. Ten consecutive deviation labels were placed on samples lined up with the

layers to examine the relationship between the extruded layers and the oscilla-

tions. Several series of points were identified and analysed on various samples

but only a typical one is shown here due to space limitations. An example of

the strain oscillations observed in subsequent layers, after using the ❢✐♥❞♣❡❛❦s

function, is shown in Figure A.12. MATLAB analysis was performed on this se-

ries of points using the techniques described in section A.2.2 and the results are

presented in the next section.

Wave properties using MATLAB

After removal of the linear trend the peaks of the waves were identified, as

shown in Figure A.12, and used to calculate the mean amplitude Ā and period

T̄ per layer, which are tabulated in Table A.2. The mean period was computed

by measuring the distance between consecutive peaks while the mean amplitude

was computed by taking the average of the modulus of the y-coordinates of each

peak and trough.

As can be seen, the average of the amplitudes for each point have very large

194
 (a) Layer 1. (b) Layer 2.

(c) Layer 3. (d) Layer 4.

(e) Layer 5. (f) Layer 6.

(g) Layer 7. (h) Layer 8.

(i) Layer 9. (j) Layer 10.

195
Figure A.12: Strain-time plots for strain oscillations observed in 10 adjacent layers.
 Table A.2: Mean amplitudes and periods of strain oscillations in each layer of a Type I coupon.

Layer Ā s Ā T̄ (sec) s T̄ (sec)

1 0.0011 0.0004 13.54 2.35
2 0.0011 0.0005 13.48 2.00
3 0.0012 0.0007 13.33 2.68
4 0.0008 0.0006 13.21 3.04
5 0.0009 0.0005 13.48 2.80
6 0.0012 0.0006 13.46 2.52
7 0.0013 0.0006 13.45 2.16
8 0.0011 0.0005 13.46 2.23
9 0.0010 0.0006 13.39 2.28
10 0.0010 0.0007 13.38 2.33

standard deviations, s Ā varies from about 36% for layer 1 to about 75% for layer

4, which, referring to Figure A.12, is to be expected given the quality of the sig-

nals. The mean periods have less variation, s T̄ being in the range of 15-20% of

T̄. What can be observed from these results is that the oscillations occur rela-

tively regularly but their effects are quite random, likely due to the previously

discussed material inhomogeneities.

An apparent phase shift was noticed to occur in consecutive layers of some

samples where a few of the peaks appear to move to the left when tracking them

from point 1 to 10 in Figure A.12. Therefore, MATLAB’s CFT was then used

to attempt to extract more useful information from the data. Furthermore, a

Fourier decomposition of the signal could be performed in the hope of identi-

fying how different wave components contribute to the overall oscillations. Ta-

ble A.3 shows the values obtained from fitting the following equation to the sig-

nal data produced by the layers,

ε(t) = A1 sin( B1 t + C1 ) (A.1)

The R2 value is included to show the goodness of the fit while Figure A.13 shows

an example of what the fitted line looked like superimposed to the original sig-

nal.

196
 Table A.3: Curve fitting for 10 layers using a single sine term.

Layer A1 B1 C1 R2 value
1 0.0007679 0.4633 2.4420 0.5001
2 0.0007084 0.4635 2.3510 0.4003
3 0.0006499 0.4638 2.2350 0.2446
4 0.0003636 0.5632 1.4080 0.1493
5 0.0003391 0.5647 1.0380 0.1382
6 0.0007122 0.4623 -0.2018 0.3397
7 0.0008538 0.4628 -0.2736 0.4148
8 0.0007621 0.4630 -0.2838 0.4160
9 0.0006579 0.4634 -0.3255 0.3439
10 0.0005352 0.4643 -0.3853 0.2198
Average 0.00063501 0.48343 0.31667
s 0.0001716 0.0424 0.122859667

Figure A.13: Example of fitting Equation A.1 to the strain-time data of a layer.

197
 The R2 values show that a single sine term is not enough to generate a good

fit since the layers produce values that are at or below 0.5. The amplitude of

the oscillations A1 is the one with the highest variation, a standard deviation of

0.0001716 so no analysis of this value was performed. B1 , the frequency of the

oscillations, is slightly better with a standard deviation of only 0.0424. This indi-

cates that the oscillations between layers are comparable and lending credence

to the samples displaying a relatively periodic oscillatory behaviour. C1 is the

phase constant and the fact that it changes from layer to layer indicates that the

peaks of the oscillations are moving so there is a propagation of strain waves

throughout the samples. Perhaps due to the dampening effect of viscosity and

the interface between the layers, the oscillations are slightly slowed from one

layer to another, thus causing the phase shift.

Next the number of sine terms in the series was increased to two producing

the following equation

ε(t) = A1 sin( B1 t + C1 ) + A2 sin( B2 t + C2 ) (A.2)

With this equation the average R2 in Table A.4 is slightly higher than in Table A.3

with a smaller standard deviation, which is to be expected. Overall the fit is still

very bad and although adding more terms would make it better, no additional

information can be gleamed from this so no further analysis was performed in

this direction. Due to the fact that signal fitting is very inaccurate, most of the

analysis done in the rest of this section is more qualitative than quantitative.

Resonance

During the initial analysis of the layers some showed signs of resonance. What

is meant by that is the fact that the local strain in some parts of the samples were

larger or smaller than in adjacent layers. For example in Figure A.14 it is possible

to see that the oscillations in the highlighted layer are smaller than in the adjacent

198
 Table A.4: Curve fitting for 10 layers using two sine terms.

Layer R2 value
1 0.5459
2 0.4565
3 0.3161
4 0.2297
5 0.283
6 0.4383
7 0.5252
8 0.5351
9 0.4471
10 0.3138
Average 0.40907
s 0.114711959

layer.

It is also possible to observe how the waveform changes shape from one layer

to the other in Figure A.12. The oscillations in layer 1 start large then become

progressively smaller in the interval t ≈ 0, 80 then the reverse happens in the

interval t ≈ 80, 180, creating a bottleneck region in the interval t ≈ 80, 110. This

pattern is similar to that produced by an amplitude modulated signal and is

roughly present in subsequent layers but in layers 6 and 7 there is no distinct

bottleneck region, the oscillations having become relatively uniform. Amplitude

modulation is where a signal m(t), such as sin( Am t), is multiplied by a another

signal, for example ε(t) from Equations A.1 and A.2, with a frequency much

larger than Am . Therefore the final signal is ε(t) sin(ω0 t) [218], see Figure A.15.

By layer 10 of Figure A.12 a new pattern seems to be forming where there is a

distinct downward trend in the interval t ≈ 0, 110 followed by an upward trend

from t ≈ 110 onwards. This coupled with the example in Figure A.14 suggests

that tensile loading induces a standing wave behaviour across the sample with

each layer representing a different point of the wave. Although difficult to tell,

layers 6 and 7 in Figure A.12 are where there might be a node , since before and

after these layers the uniformity of the signal breaks down.

199
Figure A.14: Example of two adjacent layers that have different oscillation amplitudes. The plot
highlighted in red corresponds to the strain produced by the deviation label also highlighted in
red, while the plot below the highlighted one corresponds to the deviation label just below the
highlighted one.

Striations

An observation made during the course of testing was the relationship between

the oscillatory behaviour and the striated regions generated by GOM Correlate

shown in Figure A.17. The striations were found to occur in all tensile coupons,

the strain values varying from one frame to the next but the general distribution

of the striations does not change over the course of testing, indicating that they

are a consequence of the sample’s geometry. Setting deviation points in striated

regions allows the observation of the oscillatory behaviour of these regions, as

shown in Figure A.16. In Figure A.16b the red time line, which marks the current

frame of the video being shown in GOM Correlate, is at a trough while the next

point in Figure A.16c is at a peak and the final point is close to a trough. The

fact that the red line in Figure A.16d is not exactly on a trough is likely due to

an imperfection in the sample which is also why the striations are not perfectly

horizontal and instead are skewed. The fact that the points generate alternating

200
 Modulator
Am

− Am
0 π 2π 3π 4π
Carrier
Ac

− Ac
0 π 2π 3π 4π
Modulated signal
Am + Ac
Am
0
− Am
− Am − Ac
0 π 2π 3π 4π
t

Figure A.15: Analogue amplitude modulation.

oscillations suggests that the striations might be the “unit” of the oscillations, so

placing deviation labels in adjacent regions will result in the oscillations being

out of phase by π. The alternation of peak and trough in adjacent layers was

found to be generally true for all samples. No explanation for this behaviour can

be given, though, since the striations seem to be independent of orientation.

Sample (c) in Figure A.17c, from Bartolai et al., is the same as the ±45-E1

samples in this research and, as can be seen, there are no striations in the full-

field strain images that they produced. Bartolai et al. described these as strain

localizations in the loading direction and explained that they are to be expected

since they are caused by poor welding of neighbouring roads. Yet, there are clear

differences in the strain fields produced by the two experiments and there are a

number of possible reasons for this discrepancy, including different:

• material - ABS vs PLA

201
(a) Deviation points in adjacent
striation regions. (b) Point 1.

(c) Point 2.

(d) Point 3.

Figure A.16: Strain oscillations seen in GOM from deviation labels placed in regions of
alternating striations.

202
 • printers - Mendel Max 3 vs Prusa i3 MK3 and Cubicon Single 3DP-110F

• data gathering rate and testing speed - 1 Hz at 5 mm/min vs 10 Hz at 0.5

mm/min

• sample orientation - only the printer’s XY plane (see Figure 4.1) vs E1, E2,

and E3

• cameras and their setup

With the results obtained in this research it is not possible to narrow down which

factor(s) result in the difference seen in Figure A.17. The likeliest reason why the

striations are present is the material/printer combination, but it could also be any

of the other differences discussed. As has been observed in this research, faster

testing speeds result in less visible oscillations and, when coupled with lower

data gathering rates, would therefore result in less striations. Since the experi-

ments done by Bartolai et al. were at ten times the rate of those in this research,

with ten times less data gathered, this might have resulted in no visible striations.

The sample orientation likely was not a big factor since the striations are similar

in both flat and vertical orientations (Figures Figure A.17a and Figure A.17b) but

it is possible that if ABS samples were printed in the E3 orientation and tested

at a slow speed, some striations might be observed. Finally, differences in reso-

lution and distance from the samples might have resulted in no striations in the

images captured by Bartolai et al., as can be seen when comparing the images

captured in this research vis-a-vis those from their research. As previously men-

tioned, though, it is not possible to infer if one, all, or a combination of these

differences can explain the lack of striations in Bartolai et al.’s work.

Similar to Bartolai et al., Schnittker et al. also used DIC to measure the tensile

properties of plastic samples, specifically large scale tensile samples made of ABS

reinforced with 20% (by weight) glass fibres [38]. For the dimensions of the sam-

ples see Figure A.18a. Figure A.18b shows the strain fields captured before the

rupture of one of the samples and, as can be seen at the centre of the sample, the

203
 (a) ±45-E3-01. (b) ±45-E1-01.

(c) Full field strain images of strain in the

loading direction for 100% infill density
with ±45 infill with ±45 deg (left),
+30 deg / − 60 deg (middle), and
0 deg /90 deg (right) tool-path
orientations [37].

Figure A.17: Example of striations observed in this research compared to Bartolai et al. The
strain fields in (a) and (b) are not superimposed on the video of the samples.

204
patterns are similar to those seen in Figures A.17a and A.17b. Unlike in the strain

fields produced by Bartolai et al. (see Figure A.17c), which are very smooth, in

Schnittker et al.’s images the strain fields are broken up by a spotted, striation

pattern that is reminiscent of the images captured in this research, especially in

the E2 and εxx images. Schnittker et al. don’t comment on the patterns, though,

since they are not unexpected in samples that are about to rupture.

(a) Dimensions of large tensile specimen

with a thickness of 19 mm [38].

(b) DIC strain fields taken the instant before sample rupture [38].

Figure A.18

The images capture by Schnittker et al., though, are subject to the same com-

ments made on those produced by Bartolai et al. The material, printer, sample,

and testing conditions differ from those in this research therefore a direct com-

parison cannot be made. Ituarte et al. used multimaterial jetting to produce

samples and tested them using DIC to measure their mechanical properties but

produced smooth fields like Bartolai et al. [219]. There are other instances in the

literature where DIC is used but either the material is much tougher than PLA

[220, 221, 222, 223, 224], or the samples are made of lattice structures [225]. In

both cases the direct comparison to the images captured in this research is not

205
possible.

A.3.2 Analysis of modifed type I coupons in the E3 orientation

The following testing was still based on layers because the striations are irregular

and can change shape and size during testing, whereas the layers remain evenly

spaced out and regular in their features. As previously explained, these coupons

were shorter and thicker than the regular type I samples, the objective here being

to see the effects of using a stiffer sample. Initial testing showed that the oscil-

lations were still present in the modified coupons, confirming the observations

that had been made before.

The samples were tested at 0.5 mm/min until rupture. The plots in Fig-

ure A.19 are strain-time graphs produced by the type I samples and highlighted

are the regions of clear oscillations that they produced. These signals are from a

set of 10 deviation labels on adjacent layers similar to those shown in Figure A.6.

As can be seen, the oscillations are close together and appear to be more regular

in amplitude, which is a consequence of the samples being stiffer.

Larger oscillations were also observed, which are present especially in Fig-

ure A.19c. The same signal analysis as in section A.3.1 was performed so Equa-

tion A.2, the two term sine equation, was fitted in MATLAB to the de-linearised

signals, an example of which is shown in Figure A.20. Table A.5 shows the R2

values obtained from the fit and ,as with previous attempts, MATLAB was un-

able to fit a two term equation to the signal, opting instead to fit the larger oscil-

lations rather than the small ones, which is what the R2 value in Table A.5 refers

to. Compared to Table A.4 the fit is much better and with a smaller deviation,

although still not low enough to be considered satisfactory data. This indicates

that the original theory was correct and oscillations are affected by the stiffness of

the samples. The fact that the small oscillations are themselves oscillating lends

credence to the idea of a modulated signal, where the carrier wave is the one that

MATLAB found a fit for.

206
 (a)

(b)

(c)

Figure A.19: Strain oscillations produced by modified type I coupons.

207
Figure A.20: Fitting Equation A.2 on the signal produced by modified type I coupons in
MATLAB’s CFT.

Table A.5: Quality of curve fitting for modified type I coupons.

Layer R2 value
1 0.6604
2 0.6907
3 0.6717
4 0.6074
5 0.7199
6 0.7541
7 0.6048
8 0.6332
9 0.7011
10 0.7670
Average 0.68103
s 0.056539073

208
 APPENDIX B

DERIVATION OF THE SHEAR

MODULUS FROM THE TORSION
CONSTANT

The derivation starts with what is shown on the left of Figure B.1, a thin walled,

cylindrical element of length L, radius r and thickness dr which will be consid-

ered to be part of a solid rod or wire. The end area of the cylinder is

dA = 2πrdr (B.1)

A horizontal force dF is applied to the top of the cylinder causing a torque

dT = rdF (B.2)

which causes a rotation of angle θ. The top surface of the cylinder moves a dis-

209
 Figure B.1: Cylinder before torsion (left) and after torsion (right).

tance δx. The shear stress τ and strain γ are

1 dF
τ= (B.3a)
2πr dr
δx rθ
γ= = (B.3b)
L L

which give the shear modulus G

τ L dF
G= = (B.4)
γ 2πr2 θ dr

r
The top and bottom of Equation B.4 are multiplied by r and using Equa-

tion B.2 gives

L rdF L dT
G= 3
= (B.5)
2πr θ dr 2πr3 θ dr

After rearranging for dT we have

2πr3 θG
dT = dr (B.6)
L

210
then integrating over the whole cylinder gives

Z T
2πGθ r 3
Z
dT = r dr (B.7a)
0 L 0
2πGθr4
T= (B.7b)
2L

The torque can be written as T = κθ where κ is the torsion constant. Plugging

this into Equation B.7b results in an expression that gives the shear modulus in

terms of the torsion constant,

πr4
κ=G (B.8a)
2L
2L
G=κ 4 (B.8b)
πr

211
 APPENDIX C

COMPARISON OF
CALCULATED AND
EXPERIMENTAL SPRING
CONSTANTS

212
Table C.1: Calculated and measured spring constants with the shear modulus realignment.

(k calc −k exp )
Spring d (mm) D (mm) P (mm) k exp (N/m) k calc (N/m) k exp × 100
d3D20P10 3 20 10 362.728 397.841 8.8
d3D20P12_5 3 20 12.5 346.063 402.825 14.1
d3D20P15 3 20 15 319.742 406.006 21.2
d3D30P10 3 30 10 108.013 115.005 6.1
d3D30P12_5 3 30 12.5 109.365 116.576 6.2
d3D30P15 3 30 15 107.934 117.879 8.4
d3D40P10 3 40 10 40.855 47.738 14.4
d3D40P12_5 3 40 12.5 46.256 48.334 4.3
d3D40P15 3 40 15 42.036 48.863 14.0
d4D20P10 4 20 10 1122.010 1257.370 10.8
d4D20P12_5 4 20 12.5 1111.570 1273.130 12.7
d4D20P15 4 20 15 1141.920 1283.180 11.0
d4D30P10 4 30 10 350.335 363.472 3.6
d4D30P12_5 4 30 12.5 344.054 368.437 6.6
d4D30P15 4 30 15 353.480 372.555 5.1
d4D40P10 4 40 10 142.863 150.876 5.3
d4D40P12_5 4 40 12.5 154.380 152.759 -1.1
d4D40P15 4 40 15 148.643 154.432 3.7
d5D20P10 5 20 10 2778.804 3069.760 9.5
d5D20P12_5 5 20 12.5 2985.170 3108.220 4.0
d5D20P15 5 20 15 2778.804 3132.760 11.3
d5D30P10 5 30 10 892.075 887.383 -0.5
d5D30P12_5 5 30 12.5 873.143 899.504 2.9
d5D30P15 5 30 15 870.282 909.558 4.3
d5D40P10 5 40 10 343.148 368.350 6.8
d5D40P12_5 5 40 12.5 371.106 372.947 0.5
d5D40P15 5 40 15 337.879 377.032 10.4
Average 7.6
s 5.1

213
Table C.2: Calculated and measured spring constants without the shear modulus realignment.

(k calc −k exp )
Spring d (mm) D (mm) P (mm) k exp (N/m) k calc (N/m) k exp × 100
d3D20P10 3 20 10 362.728 379.066 4.3
d3D20P12_5 3 20 12.5 346.063 376.459 8.1
d3D20P15 3 20 15 319.742 373.345 14.4
d3D30P10 3 30 10 108.013 113.094 4.5
d3D30P12_5 3 30 12.5 109.365 112.742 3.0
d3D30P15 3 30 15 107.934 112.316 3.9
d3D40P10 3 40 10 40.8554 47.8284 14.6
d3D40P12_5 3 40 12.5 46.25605 47.7439 3.1
d3D40P15 3 40 15 42.0355 47.6414 11.8
d4D20P10 4 20 10 1122.01 1198.03 6.3
d4D20P12_5 4 20 12.5 1111.57 1189.8 6.6
d4D20P15 4 20 15 1141.92 1179.95 3.2
d4D30P10 4 30 10 350.335 357.434 2.0
d4D30P12_5 4 30 12.5 344.054 356.321 3.4
d4D30P15 4 30 15 353.48 354.973 0.4
d4D40P10 4 40 10 142.863 151.161 5.5
d4D40P12_5 4 40 12.5 154.38 150.894 -2.3
d4D40P15 4 40 15 148.643 150.57 1.3
d5D20P10 5 20 10 2778.804 2924.89 5.0
d5D20P12_5 5 20 12.5 2985.17 2904.78 -2.8
d5D20P15 5 20 15 2778.804 2880.75 3.5
d5D30P10 5 30 10 892.075 872.643 -2.2
d5D30P12_5 5 30 12.5 873.143 869.923 -0.4
d5D30P15 5 30 15 870.282 866.634 -0.4
d5D40P10 5 40 10 343.148 369.046 7.0
d5D40P12_5 5 40 12.5 371.106 368.395 -0.7
d5D40P15 5 40 15 337.879 367.603 8.1
d7D20P15 7 20 15 25,009.50 24,361 2.6
Average 4.1
s 4.6

214
 PUBLICATIONS

• E. Sacco and S. K. Moon, “Additive manufacturing for space: status and

promises,” The International Journal of Advanced Manufacturing Technology,

pp. 1–24, 2019

• E. Sacco, Z. Y. Chua, and S. K. Moon, “Comparison of elastic properties for

different sample types fabricated with additive manufacturing,” in Proceed-

ings of the 3rd International Conference on Progress in Additive Manufacturing,

Singapore Centre for 3D Printing, 2018. (Pro-AM 2018)

215
 BIBLIOGRAPHY

[1] M. A. Saleh and A. E. Ragab, “Ti-6Al-4V helical spring manufacturing via

SLM: effect of geometry on shear modulus,” in Proceedings of the

International MultiConference of Engineers and Computer Scientists, vol. 2,

2013.

[2] L. He, H. Peng, M. Lin, R. Konjeti, F. Guimbretière, and J. E. Froehlich,

“Ondulé: Designing and Controlling 3D Printable Springs,” in Proceedings

of the 32nd Annual ACM Symposium on User Interface Software and

Technology, pp. 739–750, 2019.

[3] S. V. Hoa, “Development of composite springs using 4D printing

method,” Composite Structures, vol. 210, pp. 869 – 876, 2019.

[4] N. Werkheiser, “In-space manufacturing: Pioneering a sustainable path to

mars,” 2015. Presented at NASA Marshall Space Flight Center;

Huntsville, AL, United States. NASA Report/Patent Number: M15-4866.

[5] K. Rainey, “Building the future: Space station crew 3-d prints first

student-designed tool in space,” 2016. Accessed: 2016-08-01.

216
 [6] N. Werkheiser and J. Edmunson, “3D Printing in Zero-G ISS Technology

Demonstration,” 2014. Presented at 3rd Annual International Space

Station Research and Development Conference, Chicago, IL, USA. NASA

document ID: 20140011720.

[7] R. P. Hoyt, J. I. Cushing, J. T. Slostad, G. Jimmerson, T. Moser, G. Kirkos,

M. L. Jaster, and N. R. Voronka, “Spiderfab: An architecture for

self-fabricating space systems,” in AIAA Space 2013 Conference and

Exposition, p. 5509, 2013.

[8] T. McGuire, M. Hirsch, M. Parsons, S. Leake, and J. Straub, “Design for an

in-space 3d printer,” in Sensors and Systems for Space Applications IX,

vol. 9838, p. 98380V, International Society for Optics and Photonics, 2016.

[9] W. M. Marshall, M. Zemba, C. Shemelya, R. Wicker, D. Espalin,

E. MacDonald, C. Keif, and A. Kwas, “Using additive manufacturing to

print a cubesat propulsion system,” in AIAA/SAE/ASEE Joint Propulsion

Conference, NASA, 2015.

[10] P. Gradl, W. Brandsmeier, M. Calvert, S. Greene, D. O’Neal, C. Protz,

J. Richard, and K. Morgan, “Additive manufacturing overview:

Propulsion applications, design for and lessons learned,” 2017. NASA

Report/Patent Number: M17-6434.

[11] S. Masugana, “Entrepreneur seeks to boldly go where no one has gone

before: 3-d printing nearly an entire rocket,” 2018. Accessed: 2018-11-25.

[12] J. L. Finseth, “Rover nuclear rocket engine program: Overview of rover

engine tests. Final Report,” tech. rep., Sverdrup Technology, Inc.,

Huntsville, AL (United States), 1991.

[13] EOS, “Airbus Defence and Space - Additive Manufacturing process by

EOS optimizes Satellite Technology.” ✇✇✇✳❡♦s✳✐♥❢♦, 2015.

217
[14] ESA, “3D-printed deployment mechanism.” ❤tt♣s✿✴✴✇✇✇✳❡s❛✳✐♥t✴

s♣❛❝❡✐♥✐♠❛❣❡s✴■♠❛❣❡s✴✷✵✶✹✴✶✵✴✸❉✲♣r✐♥t❡❞❴❞❡♣❧♦②♠❡♥t❴♠❡❝❤❛♥✐s♠,
2014. Accessed on 2018-04-07.

[15] K. R. Gagne, D. L. Hitt, and M. R. McDevitt, “Development of an

additively manufactured microthruster for nanosatellite applications,” in

54th AIAA aerospace sciences meeting, p. 0963, 2016.

[16] Utah State University, Design and Characterization of a 3D-Printed Attitude

Control Thruster for an Interplanetary 6U CubeSat, 2016.

[17] M. Fiske and J. Edmunson, “Additive Construction with Mobile

Emplacement (ACME) 3D Printing Structures with In-Situ Resources,”

2017. Presented at the University of Alabama, Huntsville, AL, USA.

NASA Report/Patent Number: MSFC-E-DAA-TN48929.

[18] A. Meurisse, M. Sperl, and A. Makaya, “3D printing of a model building

block for a lunar base outer shell,” Materials Science and Engineering: A,

vol. 413-414, pp. 533–537, dec 2005.

[19] G. Cesaretti, E. Dini, X. D. Kestelier, V. Colla, and L. Pambaguian,

“Building components for an outpost on the lunar soil by means of a

novel 3d printing technology,” Acta Astronautica, vol. 93, pp. 430 – 450,

2014.

[20] Mechanical Engineering Community, “Types of springs.” ❤tt♣s✿

✴✴♠❡❝❤❛♥✐❝❛❧✲❡♥❣❣✳❝♦♠✴❣❛❧❧❡r②✴✐♠❛❣❡✴✷✼✺✶✲t②♣❡s✲♦❢✲s♣r✐♥❣s✴.
Accessed on 2019-01-20.

[21] S. Arestie, E. G. Lightsey, and B. Hudson, “Development of a modular,

cold gas propulsion system for small satellite applications,” Journal of

Small Satellites, vol. 1, no. 2, pp. 63–74, 2012.

218
[22] Acxess Spring, “Non-Linear Springs.”

❤tt♣s✿✴✴✇✇✇✳❛❝①❡ss♣r✐♥❣✳❝♦♠✴♥♦♥✲❧✐♥❡❛r✲s♣r✐♥❣s✳❤t♠❧. Accessed on
2019-05-15.

[23] M. Sakovsky, S. Pellegrino, and J. Costantine, “Rapid Design of

Deployable Antennas for CubeSats: A tool to help designers compare and

select antenna topologies.,” IEEE Antennas and Propagation Magazine,

vol. 59, no. 2, pp. 50–58, 2017.

[24] G. Olson, S. Pellegrino, J. Banik, and J. Costantine, “Deployable helical

antennas for CubeSats,” in 54th AIAA/ASME/ASCE/AHS/ASC Structures,

Structural Dynamics, and Materials Conference, p. 1671, 2013.

[25] ASTM International, “ASTM52921-13, Standard Terminology for Additive

Manufacturing-Coordinate Systems and Test Methodologies,” 2013.

[26] O. A. Mohamed, S. H. Masood, and J. L. Bhowmik, “Optimization of

fused deposition modeling process parameters: a review of current

research and future prospects,” Advances in Manufacturing, vol. 3, no. 1,

pp. 42–53, 2015.

[27] O. S. Es-Said, J. Foyos, R. Noorani, M. Mendelson, R. Marloth, and B. A.

Pregger, “Effect of layer orientation on mechanical properties of rapid

prototyped samples,” Materials and Manufacturing Processes, vol. 15, no. 1,

pp. 107–122, 2000.

[28] B. Rankouhi, S. Javadpour, F. Delfanian, and T. Letcher, “Failure analysis

and mechanical characterization of 3d printed abs with respect to layer

thickness and orientation,” Journal of Failure Analysis and Prevention,

vol. 16, no. 3, pp. 467–481, 2016.

[29] ASTM International, “ASTM D5379/D5379M - 12 Standard Test Method

for Shear Properties of Composite Materials by the V-Notched Beam

Method,” 2012.

219
[30] ASTM International, “ASTM D638 - 14 standard test methods for tensile

properties of plastics,” 2014.

[31] T. Letcher and M. Waytashek, “Material property testing of 3D-printed

specimen in PLA on an entry-level 3D printer,” in ASME 2014

international mechanical engineering congress and exposition, American

Society of Mechanical Engineers Digital Collection, 2014.

[32] J. Prusa. ❤tt♣s✿✴✴✇✇✇✳♣r✉s❛✸❞✳❝♦♠✴.

[33] R. W. Landgraf and R. C. Francis, “Material and processing effects on

fatigue performance of leaf springs,” SAE Transactions, pp. 1485–1494,

1979.

[34] M. Berzal, C. Barajas, D. del Mazo, J. Caja, and P. Maresca, “Simple filling

patterns to model mechanical behaviour of FDM’s test pieces under

torsion,” Procedia Manufacturing, vol. 13, pp. 786 – 793, 2017.

Manufacturing Engineering Society International Conference 2017,

MESIC 2017, 28-30 June 2017, Vigo (Pontevedra), Spain.

[35] Newcomb Spring Corp., “Rectangular wire compression springs.”

❤tt♣s✿✴✴✇✇✇✳♥❡✇❝♦♠❜s♣r✐♥❣✳❝♦♠✴r❡s♦✉r❝❡s✴
❝♦♠♣r❡ss✐♦♥✲s♣r✐♥❣s✲r❡❝t❛♥❣✉❧❛r✲✇✐r❡. Date accessed: 2018-12-10.

[36] S. A. Nama, “Modeling and analysis of a helical machined springs,” Iraqi

journal of mechanical and material engineering, vol. 15, no. 2, pp. 152–163,

2015.

[37] J. Bartolai, T. W. Simpson, and R. Xie, “Predicting strength of additively

manufactured thermoplastic polymer parts produced using material

extrusion,” Rapid Prototyping Journal, vol. 24, no. 2, pp. 321–332, 2018.

[38] K. Schnittker, E. Arrieta, X. Jimenez, D. Espalin, R. B. Wicker, and D. A.

Roberson, “Integrating digital image correlation in mechanical testing for

220
 the materials characterization of big area additive manufacturing

feedstock,” Additive Manufacturing, vol. 26, pp. 129–137, 2019.

[39] D. L. Skran, “Battle of the collossi: Sls vs falcon heavy,” 2015. Accessed:

2016-03-16.

[40] Made In Space, “Additive manufacturing facility,” 2018. Accessed:

2018-05-31.

[41] T. J. Prater, Q. A. Bean, R. D. Beshears, T. D. Rolin, N. J. Werkheiser, E. A.

Ordonez, R. M. Ryan, and F. E. Ledbetter III, “Summary report on phase I

results from the 3D printing in zero g technology demonstration mission,

volume I,” tech. rep., NASA Marshall Space Flight Center, 2016.

[42] D. Thomas, M. P. Snyder, M. Napoli, E. R. Joyce, P. Shestople, and

T. Letcher, “Effect of Acrylonitrile Butadiene Styrene Melt Extrusion

Additive Manufacturing on Mechanical Performance in Reduced

Gravity,” in AIAA SPACE and Astronautics Forum and Exposition, p. 5278,

2017.

[43] N. Werkheiser, “Nasa additive manufacturing overview,” 2017. Presented

at Military Additive Manufacturing Summit 2017, Tampa, FL, USA.

NASA Report/Patent Number: MSFC-E-DAA-TN38811.

[44] E. Ordonez, J. Edmunson, M. Fiske, E. Christiansen, J. Miller, B. A. Davis,

J. Read, M. Johnston, and J. Fikes, “Hypervelocity impact testing of

materials for additive construction: Applications on earth, the moon, and

mars,” Procedia Engineering, vol. 204, pp. 390–396, 2017.

[45] E. Atzeni, L. Iuliano, P. Minetola, and A. Salmi, “Redesign and cost

estimation of rapid manufactured plastic parts,” Rapid Prototyping Journal,

2010.

221
[46] H. Bikas, P. Stavropoulos, and G. Chryssolouris, “Additive manufacturing

methods and modelling approaches: a critical review.,” International

Journal of Advanced Manufacturing Technology, vol. 83, 2016.

[47] H. Kodama, “Automatic method for fabricating a three-dimensional

plastic model with photo-hardening polymer,” Review of scientific

instruments, vol. 52, no. 11, pp. 1770–1773, 1981.

[48] A. Khorasani, I. Gibson, M. Goldberg, and G. Littlefair, “On the role of

different annealing heat treatments on mechanical properties and

microstructure of selective laser melted and conventional wrought

ti-6al-4v,” Rapid Prototyping Journal, vol. 23, no. 2, pp. 295–304, 2017.

[49] S. L. Sing, W. Y. Yeong, F. E. Wiria, B. Y. Tay, Z. Zhao, L. Zhao, Z. Tian, and

S. Yang, “Direct selective laser sintering and melting of ceramics: a

review,” Rapid Prototyping Journal, vol. 23, no. 3, pp. 611–623, 2017.

[50] X. Wang, M. Jiang, Z. Zhou, J. Gou, and D. Hui, “3D printing of polymer

matrix composites: A review and prospective,” Composites Part B:

Engineering, vol. 110, pp. 442 – 458, 2017.

[51] C. Bang Pham, K. Fai Leong, T. Chiun Lim, and K. Sin Chian, “Rapid

freeze prototyping technique in bio-plotters for tissue scaffold

fabrication,” Rapid Prototyping Journal, vol. 14, no. 4, pp. 246–253, 2008.

[52] D. M. Atyam and N. H. Nguyen, “Designing and testing liquid engines

for additive manufacturing,” in 51st AIAA/SAE/ASEE Joint Propulsion

Conference, p. 4051, 2015.

[53] M. B. Koelbl, “Technology development and trends: Liquid rocket

propulsion,” 27-29 July 2015. Presented at 51st AIAA/SAE/ASEE Joint

Propulsion Conference (AIAA Propulsion and Energy), Orlando, FL,

USA. NASA Report/Patent Number: M15-4789.

222
[54] T. Ghidini, “An overview of current am activities at the european space

agency,” 2013. Presented at the 3D Printing & Additive Manufacturing -

Industrial Applications Global Summit 2013, London, UK.

[55] A. K. Misra, J. E. Grady, and R. Carter, “Additive manufacturing of

aerospace propulsion components,” 01 October 2015. Presented at

Additive Manufacturing for Small Manufacturers, Pittsburgh, PA, USA.

NASA Report/Patent Number: GRC-E-DAA-TN27123.

[56] M. C. Halbig and M. Singh, “Additive manufacturing of sic-based

ceramics and ceramic matrix composites,” 14 June 2015. Presented at 11th

International Conference on Ceramic Materials and Components for

Energy and Environmental Applications in Vancouver, BC, Canada.

NASA Report/Patent Number: GRC-E-DAA-TN24157.

[57] W. E. Frazier, “Metal additive manufacturing: A review,” Journal of

Materials Engineering and Performance, vol. 23, pp. 1917–1928, Jun 2014.

[58] V. K. Ball, L. B. Roberson, G. W. O’Connor, S. Trigwell, S. Bose, and

A. Bandyopadhyay, “First demonstration on direct laser fabrication of

lunar regolith parts,” Rapid Prototyping Journal, vol. 18, no. 6, pp. 451–457,

2012.

[59] N. R. Council, 3D Printing in Space. Washington, DC: The National

Academies Press, 2014.

[60] G. P. Merrill, A treatise on rocks, rock-weathering and soils. Macmillan, 1897.

[61] M. Mobrem and D. S. Adams, “Deployment analysis of the lenticular

jointed antennas onboard the mars express spacecraft,” Journal of

Spacecraft and Rockets, vol. 46, no. 2, pp. 394–402, 2009.

[62] Ultimaker, “Ultimaker Tough PLA.” Accessed: 2020-01-06.

223
[63] E. R. Ponslet and W. O. Miller, “Coil springs with constrained-layer

viscoelastic damping for passive isolation,” in Smart Structures and

Materials 1998: Passive Damping and Isolation (L. P. Davis, ed.), vol. 3327,

pp. 432 – 443, International Society for Optics and Photonics, SPIE, 1998.

[64] T. Yokota, T. Taguchi, and M. Gen, “A solution method for optimal weight

design problem of herical spring using genetic algorithms,” Computers &

Industrial Engineering, vol. 33, no. 1, pp. 71 – 76, 1997. Proceedings of the

21st International Conference on Computers and Industrial Engineering.

[65] J. C. Hendry and C. Probert, “Carbon fibre coil springs,” Materials &

Design, vol. 7, no. 6, pp. 330 – 337, 1986.

[66] G. Mastinu, M. Gobbi, and C. Miano, Optimal design of complex mechanical

systems: with applications to vehicle engineering. Springer Science & Business

Media, 2007.

[67] E. G. Merriam, J. E. Jones, and L. L. Howell, “Design of 3D-printed

titanium compliant mechanisms,” 2014.

[68] Y. Zhou, W. M. Huang, S. F. Kang, X. L. Wu, H. B. Lu, J. Fu, and H. Cui,

“From 3D to 4D printing: approaches and typical applications,” Journal of

Mechanical Science and Technology, vol. 29, no. 10, pp. 4281–4288, 2015.

[69] A. Garaigordobil, R. Ansola, E. Veguería, and I. Fernandez, “Overhang

constraint for topology optimization of self-supported compliant

mechanisms considering additive manufacturing,” Computer-Aided

Design, vol. 109, pp. 33 – 48, 2019.

[70] C. R. de Lima and G. H. Paulino, “Auxetic structure design using

compliant mechanisms: a topology optimization approach with

polygonal finite elements,” Advances in Engineering Software, vol. 129,

pp. 69 – 80, 2019.

224
[71] R. Mutlu, G. Alici, M. in het Panhuis, and G. Spinks, “Effect of flexure

hinge type on a 3D printed fully compliant prosthetic finger,” in 2015

IEEE International Conference on Advanced Intelligent Mechatronics (AIM),

pp. 790–795, IEEE, 2015.

[72] N. Meisel, A. Gaynor, C. Williams, and J. Guest, “Multiple-material

topology optimization of compliant mechanisms created via polyjet 3d

printing,” in 24th Annual international solid freeform fabrication symposium

an additive manufacturing conference, p. 28, 2013.

[73] Z. Pan, D. Ding, B. Wu, D. Cuiuri, H. Li, and J. Norrish, “Arc welding

processes for additive manufacturing: A review,” in Transactions on

Intelligent Welding Manufacturing (S. Chen, Y. Zhang, and Z. Feng, eds.),

(Singapore), pp. 3–24, Springer Singapore, 2018.

[74] Y. Kok, X. P. Tan, P. Wang, M. L. S. Nai, N. H. Loh, E. Liu, and S. B. Tor,

“Anisotropy and heterogeneity of microstructure and mechanical

properties in metal additive manufacturing: A critical review,” Materials

& Design, vol. 139, pp. 565 – 586, feb 2018.

[75] P. Parandoush and D. Lin, “A review on additive manufacturing of

polymer-fiber composites,” Composite Structures, vol. 182, pp. 36 – 53, dec

2017.

[76] B. Schoinochoritis, D. Chantzis, and K. Salonitis, “Simulation of metallic

powder bed additive manufacturing processes with the finite element

method: A critical review,” Proceedings of the Institution of Mechanical

Engineers, Part B: Journal of Engineering Manufacture, vol. 231, pp. 96–117,

aug 2017.

[77] G. Costabile, M. Fera, F. Fruggiero, A. Lambiase, and D. Pham, “Cost

models of additive manufacturing: A literature review,” International

225
 Journal of Industrial Engineering Computations, vol. 8, no. 2, pp. 263–283,

2017.

[78] A. Busachi, J. Erkoyuncu, P. Colegrove, F. Martina, C. Watts, and R. Drake,

“A review of additive manufacturing technology and cost estimation

techniques for the defence sector,” CIRP Journal of Manufacturing Science

and Technology, vol. 19, pp. 117 – 128, nov 2017.

[79] J. R. Dennison, “An overview of the dynamic interplay between the space

environment and spacecraft materials,” 2015. Laboratory of Spacecraft

Environment Interaction Engineering.

[80] N. Werkheiser, “Overview of nasa initiatives in 3d printing and additive

manufacturing,” 2014. Presented at 2014 DoD Maintenance Symposium,

Birmingham, AL, USA. NASA Report/Patent Number: M15-4252.

[81] D. C. Brown, “Nasa to launch new science mission to asteroid in 2016,”

2011. Accessed: 2015-11-23.

[82] GlobalCom Satellite Phones, “The cost of building and launching a

satellite,” 2015. Accessed: 2016-06-15.

[83] T. K. Imken, T. H. Stevenson, and E. G. Lightsey, “Design and testing of a

cold gas thruster for an interplanetary cubesat mission,” Journal of Small

Satellites, vol. 4, no. 2, pp. 371–386, 2015.

[84] M. Molitch-Hou, “3d printing has its place in martian settlements,

according to nasa,” 2016. Accessed: 2017-09-01.

[85] J. Y. Wong and A. C. Pfahnl, “3D Printing of Surgical Instruments for

Long-Duration Space Missions,” Aviation, Space, and Environmental

Medicine, vol. 85, pp. 758–763, jul 2014.

[86] Made In Space, “3d printing in zero gravity experiment,” 2018. Accessed:

2016-05-10.

226
[87] ASTM International, “ASTM D638-14 Standard Test Method for Tensile

Properties of Plastics,” 2014.

[88] ASTM International, “ASTM D790 - 15e2 Standard Test Methods for

Flexural Properties of Unreinforced and Reinforced Plastics and Electrical

Insulating Materials,” 2015.

[89] ASTM International, “ASTM D695 - 15 Standard Test Method for

Compressive Properties of Rigid Plastics,” 2015.

[90] Q. A. Bean, K. G. Cooper, J. E. Edmunson, M. M. Johnston, and M. J.

Werkheiser, “International space station (iss) 3d printer performance and

material characterization methodology,” 2015. Presented at 62nd

JANNAF Propulsion Meeting, Nashville, TN, USA. NASA Report/Patent

Number: M15-4563.

[91] R. G. Clinton and K. Morgan, “Additive manufacturing at nasa marshall

space flight center: In-space and for-space initiatives,” 2015. Presented at

the Additive Manufacturing for Defense and Government Symposium in

Huntsville, AL, USA. NASA Report/Patent Number: M16-5001.

[92] A. Boyle, “Tethers unlimited delivers 3-d printer and recycler combo to

nasa for space station,” 2018. Accessed: 2018-10-21.

[93] SBIR and NASA, “Positrusion filament recycling system for iss,” 2014.

Accessed: 2017-08-17.

[94] T. U. Inc., “Positrusion Filament Recycler, Superior Quality Filament For

3D Printers.” Brochure, 2015.

[95] A. Boyle, “Tethers Unlimited wins NASA grant to work on future FabLab

for 3-D printing in space.” ✇✇✇✳❣❡❡❦✇✐r❡✳❝♦♠, 2017.

[96] Tethers Unlimited, “Spiderfab,” 2014. Accessed: 2018-03-06.

227
 [97] S. Patane, E. R. Joyce, M. P. Snyder, and P. Shestople, “Archinaut: In-space

manufacturing and assembly for next-generation space habitats,” in AIAA

SPACE and Astronautics Forum and Exposition, p. 5227, 2017.

[98] R. G. Clinton, “Additive manufacturing for human space exploration,”

2018. Presented at the Additive Manufacturing for Aerospace and Space

conference, Munich, Germany. NASA Report/Patent Number:

MSFC-E-DAA-TN52182.

[99] J. P. Kruth, M. C. Leu, and T. Nakagawa, “Progress in Additive

Manufacturing and Rapid Prototyping,” {CIRP} Annals - Manufacturing

Technology, vol. 47, no. 2, pp. 525 – 540, 1998.

[100] G. Konecny, “Small satellites - A tool for Earth observation?,” in XXth

ISPRS Congress, Commission, vol. 4, pp. 12–23, 2004.

[101] SaRC, “Satellite Research Centre (SaRC) factsheet,” 2016.

[102] E. Mahoney, “First cubesat built by an elementary school deployed into

space,” May 2016. Accessed: 2016-12-03.

[103] amsat-uk, “Ossi-1 amateur radio cubesat launched,” April 2013.

Accessed: 2018-02-26.

[104] J. Tylka, “Evaluation of additively manufactured metals for use on oxygen

systems,” 2016. Presented at Additive Manufacturing for Propulsion

Applications, Huntsville, AL, USA. NASA Report/Patent Number:

JSC-CN-37294.

[105] R. G. Clinton, “Overview of additive manufacturing initiatives at nasa

marshall space flight center - in space and rocket engines,” Feb. 22 2017.

Presented at the Additive Manufacturing for Aerospace, Defence & Space

Conference, London, UK. NASA Report/Patent Number:

MSFC-E-DAA-TN38745.

228
[106] R. Carter, S. Draper, I. Locci, B. Lerch, D. Ellis, P. Senick, M. Meyer, J. Free,

K. Cooper, and Z. Jones, “Materials characterization of additively

manufactured components for rocket propulsion,” in 66th International

Astronautical Congress, 2015.

[107] A. Hinojos, J. Mireles, A. Reichardt, P. Frigola, P. Hosemann, L. E. Murr,

and R. B. Wicker, “Joining of inconel 718 and 316 stainless steel using

electron beam melting additive manufacturing technology,” Materials &

Design, vol. 94, pp. 17 – 27, 2016.

[108] R. W. Carter, “Grc metal additive manufacturing,” July 2015. Presented in

Cleveland, OH, USA. NASA Report/Patent Number:

GRC-E-DAA-TN25102.

[109] D. Susan, L. Brad, R. Richard, M. Richard, L. Ivan, and G. Anita, Materials

Characterization of Electron Beam Melted Ti-6Al-4V, ch. 242, pp. 1433–1440.

Wiley-Blackwell, 2016.

[110] D. L. Ellis, “Grcop-84: A high-temperature copper alloy for high-heat-flux

applications,” tech. rep., NASA Glenn Research Center, 2005. NASA

Report/Patent Number: NASA/TM-2005-213566, E-15011.

[111] D. Greitemeier, F. Palm, F. Syassen, and T. Melz, “Fatigue performance of

additive manufactured tial6v4 using electron and laser beam melting,”

International Journal of Fatigue, vol. 94, pp. 211 – 217, 2017. Fatigue and

Fracture Behavior of Additive Manufactured Parts.

[112] R. Minneci, C. Rawn, J. Bunn, J. Floyd, and Z. Jones, “Preliminary residual

stress mapping of grcop-84 fabricated by slm,” 2017. Presented at 2017

Joint Nanoscience and Neutron Scattering User Meeting, Oak Ridge, TN,

USA. NASA Report/Patent Number: M17-6172.

[113] J. Vickers, “Low cost upper stage-class propulsion (lcusp),” techreport,

NASA, 2015.

229
[114] D. Ding, Z. Pan, D. Cuiuri, and H. Li, “Wire-feed additive manufacturing

of metal components: technologies, developments and future interests,”

The International Journal of Advanced Manufacturing Technology, vol. 81,

pp. 465–481, Oct 2015.

[115] C. K. Sudbrack, M. M. Kirka, R. R. Dehoff, R. W. Carter, S. L. Semiatin,

and T. P. Gabb, “Nasa/ornl/afrl project work on ebm lshr: Additive

manufacturing of high-temperature gamma-prime strengthened ni-based

superalloys,” tech. rep., NASA Glenn Research Center, jun 2016.

[116] D. Mosher, “Defectors from SpaceX, Blue Origin, and Tesla are

developing a remarkable technology called ‘Stargate’ to help colonize

other planets,” 2018. Accessed: 2018-11-25.

[117] Relativity Space, “Terran 1,” 2018. Accessed: 2018-11-25.

[118] S. M. George, “Atomic layer deposition: an overview,” Chemical reviews,

vol. 110, no. 1, pp. 111–131, 2009.

[119] O. Mireles, C. Garcia, and Z. Jones, “Potential for additive manufacture in

nuclear thermal propulsion (ntp),” 24 February 2016. Presented at

Nuclear and Emerging Technologies for Space, Huntsville, AL, USA.

NASA Report/Patent Number: M16-5074.

[120] Community Research and Development Information Service (CORDIS),

“Amaze report summary,” 2018.

[121] N. Adkins, N. Lavery, K. Cabannes, F. Montredon, and K.Schmidtke,

“Apod 15 – invar space parts,” 2017. Presented at the AMAZE Project

Technology Forum, Coventry, UK.

[122] F. Montredon, “Sun sensor and antenna support,” 2017. Presented at the

AMAZE Project Technology Forum, Coventry, UK.

230
[123] SpaceX, “Spacex launches 3d-printed part to space, creates printed engine

chamber,” 2014. Accessed: 2016-02-02.

[124] Airbus Defence and Space, “Airbus Defence and Space optimising

components using 3D printing for new Eurostar E3000 satellite

platforms.” ✇✇✇✳❛✐r❜✉s❞❡❢❡♥❝❡❛♥❞s♣❛❝❡✳❝♦♠, 2015.

[125] M. E. Orme, M. Gschweitl, M. Ferrari, R. Vernon, I. J. Madera, R. Yancey,

and F. Mouriaux, “Additive manufacturing of lightweight, optimized,

metallic components suitable for space flight,” Journal of Spacecraft and

Rockets, vol. 54, no. 5, pp. 1050–1059, 2017.

[126] M. Orme, I. Madera, M. Gschweitl, and M. Ferrari, “Topology

optimization for additive manufacturing as an enabler for light weight

flight hardware,” Designs, vol. 2, no. 4, p. 51, 2018.

[127] M. E. Orme, M. Gschweitl, R. Vernon, M. Ferrari, I. J. Madera, R. Yancey,

and F. Mouriaux, “A demonstration of additive manufacturing as an

enabling technology for rapid satellite design and fabrication,” in

International SAMPE Technical Conference, Long Beach, CA, May, pp. 23–26,

2016.

[128] M. E. Orme, M. Gschweitl, M. Ferrari, R. Vernon, I. J. Madera, R. Yancey,

and F. Mouriaux, “Additive manufacturing of lightweight, optimized,

metallic components suitable for space flight,” Journal of Spacecraft and

Rockets, vol. 54, no. 5, pp. 1050–1059, 2017.

[129] A. Lin, “Engines - seds ucs.”

❤tt♣s✿✴✴✇✇✇✳s❡❞s✉❝s❞✳♦r❣✴♣r♦❥❡❝ts✴❡♥❣✐♥❡s✴, 2018. Accessed:

2017-09-23.

[130] J. J. Catina and K. Castonguay, Use of Additive Manufacturing to Model and

Develop Advanced Liquid Propulsion Designs. Propulsion and Energy

Forum, American Institute of Aeronautics and Astronautics, 2015.

231
[131] L. Kuentz, A. Salem, M. Singh, M. C. Halbig, and J. A. Salem, “Additive

manufacturing and characterization of polylactic acid (pla) composites

containing metal reinforcements,” 2016. Presented at the 40th

International Conference and Expo on Advanced Ceramics and

Composites, Daytona Beach, FL. USA. NASA Report/Patent Number:

GRC-E-DAA-TN29146.

[132] A. Mehrparvar, D. Pignatelli, J. Carnahan, R. Munakat, W. Lan,

A. Toorian, A. Hutputanasin, and S. Lee, “Cubesat design specification

rev 13,” The CubeSat Program, Cal Poly San Luis Obispo, US, 2014.

[133] E. A. Slejko, E. Longato, S. Seriani, P. Gallina, V. Lughi, F. Bozzolan,

A. Cuttin, M. Fragiacomo, and A. Gregorio, “Bricsat: A new satellite

structure concept,” in The 4S Symposium, 2018.

[134] C. Wyman, “Stratasys 3d printing keeps nasa satellite on time and on

budget – fdm strong enough for space,” 2015. Accessed: 2015-12-12.

[135] R. P. Mueller, I. I. Townsend, G. J. Tamasy, C. J. Evers, L. J. Sibille, J. E.

Edmunson, M. R. Fiske, J. C. Fikes, and M. Case, “Additive construction

with mobile emplacement (acme)/automated construction of

expeditionary structures (aces) materials delivery system (mds),” 2018.

Presented at the ASCE Earth and Space Conference 2018, Cleveland, OH,

USA. NASA Report/Patent Number: KSC-E-DAA-TN48594.

[136] B. Khoshnevis, D. Hwang, K.-T. Yao, and Z. Yeh, “Mega-scale fabrication

by contour crafting,” International Journal of Industrial and Systems

Engineering, vol. 1, no. 3, pp. 301–320, 2006.

[137] J. Edmunson, M. Fiske, R. Mueller, H. Alkhateb, A. Alhnoukh, H. Morris,

V. Townsend, J. Fikes, and M. Johnston, “Additive construction with

mobile emplacement: Multifaceted planetary construction materials

development,” 2018. Presented at the American Society of Civil Engineers

232
 (ASCE) Earth and Space 2018 Conference, Cleveland, OH, USA. NASA

Report/Patent Number: MSFC-E-DAA-TN54927.

[138] M. J. Werkheiser, M. Fiske, J. Edmunson, and B. Khoshnevis, “On the

development of additive construction technologies for application to

development of lunar/martian surface structures using in-situ materials,”

in AIAA SPACE 2015 Conference and Exposition, p. 4451, 2015.

[139] V. Colla and E. Dini, “Large scale 3D printing: From deep sea to the

moon,” Low-Cost 3D Printing, for Science, Education & Sustainable

Development, pp. 127–132, 2013.

[140] A. Goulas, J. G. Binner, D. S. Engstrøm, R. A. Harris, and R. J. Friel,

“Mechanical behaviour of additively manufactured lunar regolith

simulant components,” Proceedings of the Institution of Mechanical

Engineers, Part L: Journal of Materials: Design and Applications, vol. 0, no. 0,

p. 1464420718777932, 0.

[141] A. E. Jakus, K. D. Koube, N. R. Geisendorfer, and R. N. Shah, “Robust and

elastic lunar and martian structures from 3d-printed regolith inks,”

Scientific reports, vol. 7, p. 44931, 2017.

[142] A. Boley and L. Fladeland, “The Sustainable Development of Space: on

the possibility of debris stream formation during asteroid mining

operations,” in AAS/Division for Planetary Sciences Meeting Abstracts #50,

vol. 50 of AAS/Division for Planetary Sciences Meeting Abstracts, p. 119.03,

Oct. 2018.

[143] K. Lietaert, L. Thijs, B. Neirinck, T. Lapauw, B. Morrison, C. Lewicki, and

J. V. Vaerenbergh, “Meteorite as raw material for direct metal printing: A

proof of concept study,” Acta Astronautica, vol. 143, pp. 76 – 81, 2018.

[144] C. Kief, “3d printing the complete cubesat,” 2015. NASA Report/Patent

Number: NASA FS-2015-03-32-ARC.

233
[145] D. Ahn, J.-H. Kweon, J. Choi, and S. Lee, “Quantification of surface

roughness of parts processed by laminated object manufacturing,” Journal

of Materials Processing Technology, vol. 212, no. 2, pp. 339–346, 2012.

[146] A. Mostafaei, E. L. Stevens, E. T. Hughes, S. D. Biery, C. Hilla, and

M. Chmielus, “Powder bed binder jet printed alloy 625: Densification,

microstructure and mechanical properties,” Materials & Design, vol. 108,

pp. 126–135, 2016.

[147] A Fully Non-Metallic Gas Turbine Engine Enabled by Additive Manufacturing,

10 2015. Presented at 22nd International Symposium on Air Breathing

Engines (ISABE 2015), Phoenix, AZ, USA. NASA Report/Patent Number:

ISABE-2015-20168, GRC-E-DAA-TN24606.

[148] C. Dordlofva, O. Borgue, M. Panarotto, and O. Isaksson, “Drivers and

Guidelines in Design for Qualification Using Additive Manufacturing in

Space Applications,” Proceedings of the Design Society: International

Conference on Engineering Design, vol. 1, no. 1, p. 729–738, 2019.

[149] O. Borgue, M. Panarotto, and O. Isaksson, “Impact on design when

introducing additive manufacturing in space applications,” in DS 92:

Proceedings of the DESIGN 2018 15th International Design Conference,

pp. 997–1008, 2018.

[150] Function modelling and constraints replacement to support design for additive

manufacturing of satellite components, Aug 2018.

[151] O. Borgue, J. Müller, A. Leicht, M. Panarotto, and O. Isaksson, “Constraint

Replacement-Based Design for Additive Manufacturing of Satellite

Components: Ensuring Design Manufacturability through Tailored Test

Artefacts,” Aerospace, vol. 6, no. 11, p. 124, 2019.

234
[152] F. G. Bachmann, W. Hoving, Y. Lu, and K. Washio, eds., Rapid

manufacturing of lattice structures with selective laser melting, vol. 6107, SPIE,

Feb. 2006.

[153] N. Gupta, C. Weber, and S. Newsome, “Additive manufacturing: status

and opportunities,” Science and Technology Policy Institute, Washington,

2012.

[154] ASTM International, “ASTM F3122-14, Standard Guide for Evaluating

Mechanical Properties of Metal Materials Made via Additive

Manufacturing Processes,” 2014.

[155] ASTM International, “ASTM A370-16, Standard Test Methods and

Definitions for Mechanical Testing of Steel Products,” 2016.

[156] ASTM International, “ASTM B565-04(2015), Standard Test Method for

Shear Testing of Aluminum and Aluminum-Alloy Rivets and

Cold-Heading Wire and Rods,” 2015.

[157] ASTM International, “ASTM E132-04(2010), Standard Test Method for

Poisson’s Ratio at Room Temperature,” 2010.

[158] ASTM International, “ASTM E290-14, Standard Test Methods for Bend

Testing of Material for Ductility,” 2014.

[159] America Makes & ANSI Additive Manufacturing Standardization

Collaborative (AMSC), “Standardization Roadmap for Additive

Manufacturing 2.0,” tech. rep., ANSI, June 2018. Accessed: 2020-02-02.

[160] ANSI, “America Makes & ANSI Additive Manufacturing Standardization

Collaborative (AMSC).” Accessed: 2020-02-02.

[161] Marshall Space Flight Center, “MSFC-STD-3716 - standard for additively

manufactured spaceflight hardware by laser powder bed fusion in

metals,” standard, NASA, October 2017.

235
[162] Marshall Space Flight Center, “MSFC-SPEC-3717 - specification for

control and qualification of laser powder bed fusion metallurgical

processes,” standard, NASA, October 2017.

[163] K. Morgan, “Additive manufacturing overview: Propulsion applications,

design for and lessons learned,” 2017. Presented at AIAA Propulsion and

Energy Forum 2017, Atlanta, GA, USA. NASA Report/Patent Number:

MSFC-E-DAA-TN44554.

[164] R. Ranjan, S. K. Chou, F. Riaz, and K. Karthikeyan, “Cold gas micro

propulsion development for satellite application,” Energy Procedia,

vol. 143, pp. 754 – 761, 2017. Leveraging Energy Technologies and Policy

Options for Low Carbon Cities.

[165] P. Muri, O. Challa, and J. McNair, “Enhancing small satellite

communication through effective antenna system design,” in

2010-MILCOM 2010 MILITARY COMMUNICATIONS CONFERENCE,

pp. 347–352, IEEE, 2010.

[166] S.-H. Ahn, M. Montero, D. Odell, S. Roundy, and P. K. Wright,

“Anisotropic material properties of fused deposition modeling ABS,”

Rapid prototyping journal, vol. 8, no. 4, pp. 248–257, 2002.

[167] C. Ziemian, M. Sharma, and S. Ziemian, “Anisotropic mechanical

properties of ABS parts fabricated by fused deposition modelling,”

Mechanical engineering, vol. 23, 2012.

[168] S. Hassanifard and S. M. Hashemi, “On the strain-life fatigue parameters

of additive manufactured plastic materials through fused filament

fabrication process,” Additive Manufacturing, vol. 32, p. 100973, 2020.

[169] Stratasys, “Stratasys Direct Manufacturing Builds the First 3D Printed

Parts to Function on the Exterior of a Satellite.”

✇✇✇✳str❛t❛s②s❞✐r❡❝t✳❝♦♠, 2017.

236
[170] D. Popescu, A. Zapciu, C. Amza, F. Baciu, and R. Marinescu, “FDM

process parameters influence over the mechanical properties of polymer

specimens: A review,” Polymer Testing, vol. 69, pp. 157–166, 2018.

[171] O. H. Ezeh and L. Susmel, “Fatigue behaviour of additively manufactured

polylactide (PLA),” Procedia Structural Integrity, vol. 13, pp. 728–734, 2018.

[172] A. Rodríguez-Panes, J. Claver, and A. Camacho, “The influence of

manufacturing parameters on the mechanical behaviour of pla and abs

pieces manufactured by fdm: a comparative analysis,” Materials, vol. 11,

no. 8, p. 1333, 2018.

[173] S. Cheves, “A pilot study to evaluate VOCs outgassed in polymer

filaments used in 3D printing,” A proposed study for a National Student

Design Competition, 2014.

[174] F. Theodore, “Method of making plastic springs,” May 1945. US Patent

2,375,357.

[175] F. W. Reinhart, M. C. Slone, H. Leon, and D. A. George, “Reinforced

plastic springs,” Sept. 1958. US Patent 2,852,424.

[176] P. D. Cochrane, “Life extension in hard disk drives through vibration

dampening using pre-stressed polymer springs,” Nov. 2006. US Patent

7,142,419.

[177] M. Almasri, W. Zhang, A. Kine, Y. Chan, J. C. LaRue, and R. Nelson,

“Tunable infrared filter based on elastic polymer springs,” in Advanced

Sensor Technologies for Nondestructive Evaluation and Structural Health

Monitoring (N. Meyendorf, G. Y. Baaklini, and B. Michel, eds.), vol. 5770,

pp. 190 – 198, International Society for Optics and Photonics, SPIE, 2005.

[178] G. B. Jadhav, V. Gawande, and T. Ambi, “Review on development and

analysis of helical spring with combination of conventional and

237
 composite materials,” International Journal of Engineering Research and

General Science, vol. 3, no. 2, pp. 1180–1184, 2015.

[179] B. H. Lee, J. Abdullah, and Z. A. Khan, “Optimization of rapid

prototyping parameters for production of flexible ABS object,” Journal of

materials processing technology, vol. 169, no. 1, pp. 54–61, 2005.

[180] A. K. Sood, R. K. Ohdar, and S. S. Mahapatra, “Improving dimensional

accuracy of fused deposition modelling processed part using grey

Taguchi method,” Materials & Design, vol. 30, no. 10, pp. 4243–4252, 2009.

[181] A. Bagsik, V. Schöppner, and E. Klemp, “FDM part quality manufactured

with Ultem* 9085,” in 14th international scientific conference on polymeric

materials, vol. 15, pp. 307–315, 2010.

[182] F. Rayegani and G. C. Onwubolu, “Fused deposition modelling (FDM)

process parameter prediction and optimization using group method for

data handling (GMDH) and differential evolution (DE),” The International

Journal of Advanced Manufacturing Technology, vol. 73, no. 1-4, pp. 509–519,

2014.

[183] ASTM International, “ASTM52921-13, Standard Terminology for Additive

Manufacturing-Coordinate Systems and Test Methodologies,” 2013.

[184] R. Sayre III, “A comparative finite element stress analysis of isotropic and

fusion deposited 3D printed polymer,” in Submitt. to Grad. Fac. Rensselaer

Polytech. Inst. Partial Fulfillment Requir. degree MASTERS Eng, 2014.

[185] M. Domingo-Espin, J. M. Puigoriol-Forcada, A.-A. Garcia-Granada,

J. Lluma, S. Borros, and G. Reyes, “Mechanical property characterization

and simulation of fused deposition modeling polycarbonate parts,”

Materials & Design, vol. 83, pp. 670–677, 2015.

238
[186] P. Feng, X. Meng, J.-F. Chen, and L. Ye, “Mechanical properties of

structures 3D printed with cementitious powders,” Construction and

Building Materials, vol. 93, pp. 486 – 497, 2015.

[187] J. Cantrell, S. Rohde, D. Damiani, R. Gurnani, L. DiSandro, J. Anton,

A. Young, A. Jerez, D. Steinbach, C. Kroese, and P. Ifju, “Experimental

Characterization of the Mechanical Properties of 3D Printed ABS and

Polycarbonate Parts,” in Advancement of Optical Methods in Experimental

Mechanics, Volume 3: Proceedings of the 2016 Annual Conference on

Experimental and Applied Mechanics, pp. 89–105, Springer International

Publishing, 2017.

[188] C. Casavola, A. Cazzato, V. Moramarco, and C. Pappalettere, “Orthotropic

mechanical properties of fused deposition modelling parts described by

classical laminate theory,” Materials & Design, vol. 90, pp. 453 – 458, 2016.

[189] L. Li, Q. Sun, C. Bellehumeur, and P. Gu, “Composite modeling and

analysis of fdm prototypes for design and fabrication of functionally

graded parts,” in Solid Freeform Fabrication Symposium (SFF), Austin, TX,

Aug, pp. 6–8, 2001.

[190] M. Somireddy, A. Czekanski, and C. V. Singh, “Development of

constitutive material model of 3D printed structure via FDM,” Materials

Today Communications, vol. 15, pp. 143–152, 2018.

[191] J. F. Rodríguez, J. P. Thomas, and J. E. Renaud, “Mechanical behavior of

acrylonitrile butadiene styrene fused deposition materials modeling,”

Rapid Prototyping Journal, vol. 9, no. 4, pp. 219–230, 2003.

[192] ASTM International, “ASTM D4255/D4255M - 01 Standard Test Method

for In-Plane Shear Properties of Polymer Matrix Composite Materials by

the Rail Shear Method,” 2001.

239
[193] Shimadzu, “AGS-X Series Universal Electromechanical Test Frames.”

❤tt♣s✿✴✴✇✇✇✳ss✐✳s❤✐♠❛❞③✉✳❝♦♠✴♣r♦❞✉❝ts✴
✉♥✐✈❡rs❛❧✲t❡♥s✐❧❡✲t❡st✐♥❣✴❛❣s✲①✲s♣❡❝✐❢✐❝❛t✐♦♥s✳❤t♠❧, 2018.
Accessed: 2017-08-21.

[194] J. Bartolai, A. E. Wilson-Heid, J. R. Kruse, A. M. Beese, and T. W. Simpson,

“Full Field Strain Measurement of Material Extrusion Additive

Manufacturing Parts with Solid and Sparse Infill Geometries,” JOM,

vol. 71, no. 3, pp. 871–879, 2019.

[195] GOM, “Gom correlate.” ❤tt♣s✿✴✴✇✇✇✳❣♦♠✲❝♦rr❡❧❛t❡✳❝♦♠✴, 2018. Date

accessed: 2016-05-31.

[196] N. McCormick and J. Lord, “Digital Image Correlation,” Materials Today,

vol. 13, no. 12, pp. 52 – 54, 2010.

[197] P. Bai, Y. Xu, F. Zhu, and D. Lei, “A novel method to compensate

systematic errors due to undermatched shape functions in digital image

correlation,” Optics and Lasers in Engineering, vol. 126, p. 105907, 2020.

[198] S. A. Tronvoll, T. Welo, and C. W. Elverum, “The effects of voids on

structural properties of fused deposition modelled parts: a probabilistic

approach,” The International Journal of Advanced Manufacturing Technology,

vol. 97, pp. 3607–3618, Aug 2018.

[199] J. Torres, J. Cotelo, J. Karl, and A. P. Gordon, “Mechanical Property

Optimization of FDM PLA in Shear with Multiple Objectives,” JOM,

vol. 67, pp. 1183–1193, May 2015.

[200] Society of Automotive Engineers, Manual on design and application of helical

and spiral springs. New York, 1958. Retrieved from

http://hdl.handle.net/2027/coo.31924004648592.

[201] K. L. Richards, Design engineer’s handbook. CRC Press, 2012.

240
[202] Z. Marciniak, J. L. Duncan, and S. J. Hu, “6 - Bending of sheet,” in

Mechanics of Sheet Metal Forming (Second Edition) (Z. Marciniak, J. Duncan,

and S. Hu, eds.), pp. 82 – 107, Oxford: Butterworth-Heinemann, second

edition ed., 2002.

[203] E. J. Hearn, “Chapter 26 - FATIGUE, CREEP AND FRACTURE,” in

Mechanics of Materials (Second Edition) (E. HEARN, ed.), pp. 839 – 905,

Butterworth-Heinemann, second edition ed., 1985.

[204] A. M. Wahl, Mechanical springs. Penton Publishing Company, 1944.

[205] T. C. T. Ting, T. Chen, and K. S. Li, “Torsion of an isotropic shaft of

arbitrary cross-section embedded with multicoated or graded circular

cylinders of cylindrically orthotropic materials,” Quarterly Journal of

Mechanics and Applied Mathematics, vol. 57, no. 3, pp. 347–362, 2004.

[206] R. Lipton, “An isoperimetric inequality for the torsional rigidity of

imperfectly bonded fiber reinforced cylinders,” Journal of elasticity, vol. 55,

no. 1, pp. 1–10, 1999.

[207] T. Chen and C.-J. Wei, “Saint-Venant torsion of anisotropic shafts:

theoretical frameworks, extremal bounds and affine transformations,”

Quarterly Journal of Mechanics and Applied Mathematics, vol. 58, no. 2,

pp. 269–287, 2005.

[208] J. P. Smith, “Highly accurate beam torsion solutions using the p-version

finite element method,” NASA, 1996. NASA Technical Paper 3608.

[209] L. A. M. Danao and R. M. Cabrera, “Torsion of a rectangular prismatic

bar: solution using a power fit model,” Philippine Engineering Journal,

vol. 28, no. 1, pp. 77–98, 2007.

[210] O. C. Zienkiewicz, R. L. Taylor, and J. Z. Zhu, “Chapter 5 - Field Problems:

A Multidimensional Finite Element Method,” in The Finite Element

241
 Method: its Basis and Fundamentals (Seventh Edition) (O. Zienkiewicz,

R. Taylor, and J. Zhu, eds.), pp. 115 – 149, Oxford:

Butterworth-Heinemann, seventh edition ed., 2013.

[211] K. Darılmaz, E. Orakdöğen, and K. Girgin, “Saint-Venant torsion of

arbitrarily shaped orthotropic composite or FGM sections by a hybrid

finite element approach,” Acta Mechanica, vol. 229, p. 1387, Mar. 2018.

[212] S. Timoshenko and J. N. Goodier, Theory of elasticity. Engineering societies

monographs, New York : McGraw-Hill, [1969, c1970], 1969.

[213] L. S. Srinath, Advanced mechanics of solids. New Delhi : Tata McGraw-Hill,

c1980., 1980.

[214] W. S. Slaughter, The Linearized Theory of Elasticity. Boston, MA : Birkhäuser

Boston : Imprint: Birkhäuser, 2002., 2002.

[215] X. Rongqiao, H. Jiansheng, and C. Weiqiu, “Saint-Venant torsion of

orthotropic bars with inhomogeneous rectangular cross section,”

Composite Structures, vol. 92, no. 6, pp. 1449 – 1457, 2010.

[216] K. Hsieh, Numerical modeling and analysis of composite beam structures

subjected to torsional loading. PhD thesis, Virginia Tech, 2007.

[217] J. White, R. C. and N. K. Cooperrider, “Guideway-Suspension Tradeoffs

in Rail Vehicle Systems,” Journal of Dynamic Systems, Measurement, and

Control, vol. 103, pp. 237–244, 09 1981.

[218] L. F. Chaparro and A. Akan, “Chapter 2 - Continuous-Time Systems,” in

Signals and Systems Using MATLAB (Third Edition) (L. F. Chaparro and

A. Akan, eds.), pp. 115 – 165, Academic Press, third edition ed., 2019.

[219] I. F. Ituarte, N. Boddeti, V. Hassani, M. L. Dunn, and D. W. Rosen, “Design

and additive manufacture of functionally graded structures based on

digital materials,” Additive Manufacturing, vol. 30, p. 100839, 2019.

242
[220] J. L. Bartlett, B. P. Croom, J. Burdick, D. Henkel, and X. Li, “Revealing

mechanisms of residual stress development in additive manufacturing via

digital image correlation,” Additive Manufacturing, vol. 22, pp. 1 – 12, 2018.

[221] D. Foehring, H. B. Chew, and J. Lambros, “Characterizing the tensile

behavior of additively manufactured Ti-6Al-4V using multiscale digital

image correlation,” Materials Science and Engineering: A, vol. 724, pp. 536 –

546, 2018.

[222] Y. Balit, E. Charkaluk, and A. Constantinescu, “Digital image correlation

for microstructural analysis of deformation pattern in additively

manufactured 316L thin walls,” Additive Manufacturing, vol. 31, p. 100862,

2020.

[223] A. S. Wu, D. W. Brown, M. Kumar, G. F. Gallegos, and W. E. King, “An

experimental investigation into additive manufacturing-induced residual

stresses in 316L stainless steel,” Metallurgical and Materials Transactions A,

vol. 45, no. 13, pp. 6260–6270, 2014.

[224] D. Garcia, Z. Wu, J. Y. Kim, H. Z. Yu, and Y. Zhu, “Heterogeneous

materials design in additive manufacturing: Model calibration and

uncertainty-guided model selection,” Additive Manufacturing, vol. 27,

pp. 61 – 71, 2019.

[225] N. Vanderesse, A. Richter, N. Nuño, and P. Bocher, “Measurement of

deformation heterogeneities in additive manufactured lattice materials by

Digital Image Correlation: Strain maps analysis and reliability

assessment,” Journal of the Mechanical Behavior of Biomedical Materials,

vol. 86, pp. 397 – 408, 2018.

[226] E. Sacco and S. K. Moon, “Additive manufacturing for space: status and

promises,” The International Journal of Advanced Manufacturing Technology,

pp. 1–24, 2019.

243
[227] E. Sacco, Z. Y. Chua, and S. K. Moon, “Comparison of elastic properties

for different sample types fabricated with additive manufacturing,” in

Proceedings of the 3rd International Conference on Progress in Additive

Manufacturing, Singapore Centre for 3D Printing, 2018. (Pro-AM 2018).

244