You are on page 1of 114

Masterthesis

Soft ionic strain sensors


through a new multimaterial
3D printing process

Andreas Frutiger
(January - June 2014)

Advisors: Supervisors:
Prof. Dr. Conor Walsh Joseph Muth
Prof. Dr. Jennifer Lewis Daniel Vogt
Prof. Dr. Janos Vörös Dr. Yigit Menguc
Acknowledgements
Hereby I would like to express my thanks to all the people that contributed to the success of this project.
First, Joseph Muth, who was and is an extraordinary coworker and friend, he contributed with many
ideas for the printing process development. Prof. Alexandre Campo, who provided a lot of valuable input
for the readout electronics. Yigit Menguc, who encouraged me to try an ionic conducting approach for
the sensor material at the beginning of the project. Daniel Vogt, who always had the right electrical
components at hand. Kai Schmidth, Gregory Whittaker, Markus Horvath and my parents Bernhard
and Doris Frutiger for the mental support during tough phases of the project. Lori Sanders for teaching
me professional photography and helping with the pictures. Prof. Janos Vörös for the uncomplicated
supervision of this thesis from Switzerland. My principal investigators Prof. Conor Walsh and Prof.
Jennifer Lewis for the opportunity to work and be creative in such a multidisciplinary environment. The
success of this thesis would not have been possible without the collaborative e↵ort. A special thanks also
goes to all the other members of the Biodesign Lab and the Lewis Lab, it was an amazing time. I would
also like to thank the Ernst Göhner Foundation and the Swiss Study Foundation for their financial and
intellectual support over the last three years.

1
Contents

Acknowledgements 1

Abstract 13

1. Introduction 15
1.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.2. State of the art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3. Aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2. Project definition 17
2.1. Problem statement and design goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2. Specifications sheets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.1. Functional requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3. Functional Specifications sheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3.1. Temporal requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.2. Sensor requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.3. Ergonomic requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.4. Fabrication requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.5. Academic vs project requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.6. Risks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.7. Relative importance of the functional specifications for later evaluation of the de-
veloped solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4. Part functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4.1. Solutions to part functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5. Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5.1. Concept 1 - Strain sensitive fibers for joint kinematics . . . . . . . . . . . . . . . . 23
2.5.2. Concept 2 - Pressure and strain sensitive patches . . . . . . . . . . . . . . . . . . . 23
2.6. Evaluation of the developed concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3. Sensor model 27
3.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2. Model derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.1. Unelongated state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.2. Elongated state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

4. Sensor Material development 31


4.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2. Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2.1. Ink preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2.2. Recipies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.2.3. Rheological characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.2.4. Time stability assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.3.1. Rheological properties for printing . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.3.2. Time stability assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.4. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3
Contents

5. Nozzle Design 39
5.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
5.2. Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
5.2.1. Nozzle assembly process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.4. Discussion and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

6. Printing Process development 43


6.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.2. Speed tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.2.1. Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.2.2. Ecoflex 30 Speed test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.2.3. Dragonskin 10 Speed test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.3. Printing filaments for sensor fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.3.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.3.2. Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
6.3.3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
6.4. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

7. Sensor Fabrication 51
7.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
7.2. Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
7.3. Electrical connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
7.4. End cap fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
7.5. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

8. Readout 55
8.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
8.1.1. Ideal readout circuit model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
8.1.2. Strain dependence of decay time . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
8.1.3. Non-ideal sensor model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
8.2. Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
8.2.1. Hardware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
8.2.2. Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
8.3. Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
8.3.1. Accuracy assessment of readout electronics . . . . . . . . . . . . . . . . . . . . . . 65
8.4. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

9. Sensor Characterization 69
9.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
9.2. Functionality assessment of the sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
9.2.1. Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
9.2.2. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
9.2.3. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
9.3. Failure mode e↵ect analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
9.4. Static testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
9.4.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
9.4.2. Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
9.4.3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
9.4.4. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
9.5. Dynamic testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
9.5.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
9.5.2. Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
9.5.3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
9.5.4. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
9.6. Mechanical testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
9.6.1. Elongation at break . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

4
Contents

9.6.2. Connection cable strength . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84


9.7. Discussion sensor characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
9.7.1. Proposed approach for sensor optimization . . . . . . . . . . . . . . . . . . . . . . 85

10.Applications 87
10.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
10.2. Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
10.3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
10.4. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

11.Conclusion and Outlook 89


11.1. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
11.2. Significance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
11.3. Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

A. Sensing Material Development 95


A.1. Printing of Core/Shell eGaIn/Ecoflex wires . . . . . . . . . . . . . . . . . . . . . . . . . . 95

B. Nozzle Fabrication Data 96

C. Printing Process Data 99

D. Readout Electronics 103


D.1. Additional functions in the arduino file “Static readout half signal for linear fit.ino” . . . 103
D.2. File describtion for the readout version with a multiplexer and two resistors . . . . . . . . 105

E. Sensor charaterization data 107


E.1. Functionality assessment sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
E.1.1. Functionality assessment long strains . . . . . . . . . . . . . . . . . . . . . . . . . . 111
E.2. Static testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
E.3. Dynamic testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
E.3.1. Connection cable strength load/extension diagrams . . . . . . . . . . . . . . . . . . 112

5
List of Figures

2.1. Relation between surface strain and joint angle in human body movements . . . . . . . . 18
2.2. Force transmission element at the ankle . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3. Evaluation criteria for sensor solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4. Concept 1 - four core shell capacitive strain sensor with extruder nozzle. . . . . . . . . . 23
2.5. Concept 2 - Pressure and strain sensitive patch . . . . . . . . . . . . . . . . . . . . . . . . 23

3.1. Schematic cross section of the sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27


3.2. Cross section of the sensor with its equivalent circuit. . . . . . . . . . . . . . . . . . . . . 28
3.3. Equivalent circuit diagram of the sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.4. Capacitive sensitivity change with respect to the ratio of the two radii of the dielectric. . 30

4.1. Rheological behavior of the inks used for printing. . . . . . . . . . . . . . . . . . . . . . . 34


4.2. Container dependence of the cooling process on the final rheological properties of the ink. 35
4.3. The ink before water absorption and after 38 days. . . . . . . . . . . . . . . . . . . . . . 35
4.4. Change of electrical properties and weight gain of the ink over time. . . . . . . . . . . . . 36
4.5. Rheological behavior of ink before and after water absorption . . . . . . . . . . . . . . . . 36

5.1. The nozzle assembly process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40


5.2. The developed print head. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.3. Cross section through the final nozzle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

6.1. Overview printing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44


6.2. Macroscopic and microscopic images of the printed Ecoflex filaments. . . . . . . . . . . . 45
6.3. Printed filaments as a function of print speed . . . . . . . . . . . . . . . . . . . . . . . . . 46
6.4. Necking of the filament at maximum print velocity . . . . . . . . . . . . . . . . . . . . . . 46
6.5. Thickness of di↵erent layers as a function of print speed. . . . . . . . . . . . . . . . . . . . 47
6.6. Printing process of the sensor filaments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

7.1. Evolution of the fabricated sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51


7.2. Schematic illustrating the electrical connection process . . . . . . . . . . . . . . . . . . . . 52
7.3. Microscopic images illustrates the electrical connection process . . . . . . . . . . . . . . . 52
7.4. The end cap fabrication process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
7.5. Illustration of the final sensor with the electrical connections and the end cap. . . . . . . 54

8.1. Equivalent circuit of the measurement setup for the readout of the sensor properties . . . 55
8.2. Equivalent circuit model for the readout using a multiplexer . . . . . . . . . . . . . . . . . 58
8.3. Connection setup for the readout electronics. . . . . . . . . . . . . . . . . . . . . . . . . . 60
8.4. Signal flow in the readout electronics setup. . . . . . . . . . . . . . . . . . . . . . . . . . 60
8.5. The developed user interface for decay time monitoring and the linear fit of the decaying
signal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
8.6. The developed user interface for capacitance and resistance monitoring. . . . . . . . . . . 65
8.7. Accuracy assessment of the readout electronics. . . . . . . . . . . . . . . . . . . . . . . . 66
8.8. Loop rate of the readout electronics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

9.1. Sensor 2 - Displaying an open circuit response . . . . . . . . . . . . . . . . . . . . . . . . 70


9.2. Sensor 12 - Displaying a short circuit response . . . . . . . . . . . . . . . . . . . . . . . . 70
9.3. Sensor 9 - Displaying the expected response with one decay time . . . . . . . . . . . . . . 70
9.4. Sensor 32 - Displaying two decay times at higher strains . . . . . . . . . . . . . . . . . . . 71
9.5. Sensor 9 - Second decay time at larger strains . . . . . . . . . . . . . . . . . . . . . . . . . 71

7
List of Figures

9.6. First decay time remains constant as a function of strain . . . . . . . . . . . . . . . . . . . 72


9.7. Sensor 25 with only half of the linearized signal. . . . . . . . . . . . . . . . . . . . . . . . 72
9.8. Explanation of the two observed decay times by a simulation experiment . . . . . . . . . . 73
9.9. Typical sensor failures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
9.10. Experimental setup for the static characterization of the sensors . . . . . . . . . . . . . . 76
9.11. Sensor 6 Static Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
9.12. Sensor 8 Static Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
9.13. Sensor 9 Static Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
9.14. Sensor 14 Static Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
9.15. Sensor 25 Static Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
9.16. Sensor 32 Static Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
9.17. Dynamic testing setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
9.18. Capacitance signal dynamic test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
9.19. Resistance signal dynamic test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
9.20. Decay time signal dynamic test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

10.1. Walking monitoring with a strain sensor attached to the knee . . . . . . . . . . . . . . . . 87


10.2. Normalized decay time data for walking at di↵erent speeds. . . . . . . . . . . . . . . . . 88
10.3. A fully integrated sensor into a fabric that could sense the movement of the wrist. . . . . 88

A.1. eGaIn core printed into a Ecoflex® 00-30 shell. . . . . . . . . . . . . . . . . . . . . . . . . 95


A.2. Transients during the conductivity measurements . . . . . . . . . . . . . . . . . . . . . . . 95

C.1. Data acquisiton for the cross sections of the printed filaments. . . . . . . . . . . . . . . . 101
C.2. Data acquisiton for the top down images of the printed filaments. . . . . . . . . . . . . . 101

E.1. Sensor 3 Raw Signal and Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107


E.2. Sensor 6 Raw Signal and Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
E.3. Sensor 8 Raw Signal and Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
E.4. Sensor 11 Raw Signal and Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
E.5. Sensor 13 Raw Signal and Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
E.6. Sensor 14 Raw Signal and Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
E.7. Sensor 23 Raw Signal and Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
E.8. Sensor 25 Raw Signal and Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
E.9. Sensor 27 Raw Signal and Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
E.10.Sensor 30 Raw Signal and Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
E.11.Two decay times at very large strains for almost all sensors . . . . . . . . . . . . . . . . . 111
E.12.Load/Extension diagrams for the mechanical characterization of the connections. . . . . 112
E.13.Load/Strain curves for three sensors until failure. . . . . . . . . . . . . . . . . . . . . . . 113

8
List of Tables

2.1. Weighting of the di↵erent requirements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21


2.2. Solutions to part functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3. Selection of the most promising concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

4.1. Requirements for the two inks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

6.1. Ecoflex 30 print parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44


6.2. Dragonskin 10 print parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.3. Thickness and o↵-centering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.4. Print parameters for the sensor fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

9.1. Failure mode e↵ect analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74


9.2. Overview of the experimentally determined sensor quantities. . . . . . . . . . . . . . . . . 79
9.3. Overview of sensor quantities determined from the top down images. . . . . . . . . . . . 79
9.4. Measured capacitance and calculated capacitance from the top down images with the
results shifted by an amount of 3 pF. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
9.5. Actuation amplitude and frequency for the evaluated dynamic tests. . . . . . . . . . . . . 81
9.6. Mean absolute error for the capacitance readings at di↵erent amplitudes and frequencies. 83
9.7. Mean absolute error for the resistance readings at di↵erent amplitudes and frequencies. . 83
9.8. Mean absolute error for the decay time readings at di↵erent amplitudes and frequencies. 84
9.9. Values for the load and strain at break . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
9.10. Values for the pullout forces of the connections. . . . . . . . . . . . . . . . . . . . . . . . 84

B.1. Tubing sizes overview. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96


B.2. Tubing sizes of the core and shells of the farbricated nozzles . . . . . . . . . . . . . . . . 97
B.3. Tubing sizes of the intermediate layers of the fabricated nozzles . . . . . . . . . . . . . . 98

C.1. Dragonskin 10 speed test raw data from top down microscopic images . . . . . . . . . . . 99
C.2. Dragonskin speed test raw data from crosssection images. . . . . . . . . . . . . . . . . . . 100
C.3. Top down radii values for the fabricated filaments that showed no damage under the
microscope. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

E.1. The strain range over which the fit for capacitance and resistance was performed for each
sensor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
E.2. Actuation amplitude and frequency for all di↵erent dynamic tests. . . . . . . . . . . . . . 112

9
Nomenclature
ADC Analog to digital converter
eGaIn eutectic Gallium Indium alloy

GUI Graphical user interface


MEMS Micro electrical mechanical system

NBR nitrile butadien rubber

PCB printed circuit board

PDMS Polydimethylenesilane
SBS styrene-block-butadiene-block-styrene

SNR Signal to Noise ratio

11
Abstract
In this thesis we propose a novel fiber-like, textile integrated strain sensor based on a cylindrical capacitor
that has a linear response over a strain range of 250 %. The architecture of the sensor is a four layer,
core-shell cylindrical fiber. Layers 1 and 3 made of a ionic conductive fluid, and layers 2 and 4 are made
of a silicone elastomer. The structure acts like a cylindrical capacitor whose value increases linearly
with strain. A new multimaterial printing process was developed for the fabrication of the filament. We
engineered a print head that is capable of extruding a four layer, core shell filament with up to four
di↵erent materials at the same time. Furthermore, a novel printable ionic conducting fluid on the basis
of glycerol (solvent), sodium chloride (ionic conductor) and PEG 1500 (rheology modifier) was developed
for the conducting material of the sensor. The fluid exhibited the shear thinning properties required
for printing, was non-volatile, hygroscopic (water absorbtion up to 30 w%) and exhibited a conductivity
between 2-3 mS/cm. Sensors of di↵erent lengths and diameters can be fabricated through alterations in
the print parameters. For attachement of the filaments to the fabric, a sensor end cap was invented that
exhibited a gradient in material properties and cross sectional area, facilitating a smooth transition in
mechanical properties from the soft sensor body to the more rigid fabric. The electrical connection to the
sensor was enabled through two silver wires that were embedded into the end cap. Through a loop within
the elastomer, this connection was mechanically reinforced. The readout principle consisted of applying
a step response to the system and measuring the RC-decay time on a resistor in series to the sensor. An
analytical model for the step response of this readout system as a function of sensor strain was derived
and allowed for the determination of the sensors resistance and capacitance as a function of strain. This
readout principle was then implemented on an ATMEGA328P microcontroller. A bandwidth between 50
and 150 Hz was achieved. The sensors showed excellent agreement with the model in a strain range from
0-250 %, and sensor resistance and capacitance was between 0.8-4.22 M ⌦ and 3.78-20 pF , respectively.
The sensors displayed a gauge factor of up to 0.51. Applied strains with a frequency of up to 0.5 Hz
were tractable with mean average errors below 4.3 %. The sensors were strechable up to 770 % with a
force of only 3 N. Finally, we demonstrated the sensor’s use as a wearable device by integrating it into
a textile. The textile was anchored to a user’s knee and used to analyze joint kinematics at di↵erent
walking speeds.

13
1. Introduction
1.1. Motivation
Over the past 50 years, the field of embedded electronics has made a huge leap forward and has provided
us with minitiaturized circuits and sensors. These devices are great as long as they are used in a rigid
environment. However, rigid form factors are unacceptable for applications in which conformity and
extensiblity are required. Such fields include wearable electronics, soft robotics, and implantable medical
devices and soft sensors are actively being developed [1]. Although some work has been done to integrate
rigid MEMS sensors into a soft matrix [2], these devices are not truly soft and the achievable spatial
resolution is limited. Truly soft and compliant sensors that enable soft robotic devices a human like
proprioceptive and tactile information would be a huge step forward. With appropriate sensing ability,
these soft robotic devices exhibit a tremendous potential both in interfacing directly with humans and
enabling complicated or dangerous operations without requiring direct human presence. Applications
for interfacing with humans lie especially in the area of prosthetics, rehabilitation devices [3], surgery
and augmentation of human capabilities, whereas teleoperated soft robotic devices could be used for
space missions and replace human operators in chemical or nuclear worst case scenarios. A review on
these challenges in soft robotics was written by Pfeifer et al [4]. Furthermore, the food industry still
relies on a large amount of human handling. A device that could pick and place delicate objects like
eggs or fruits could increase the efficiency of these factories [5]. In order to perform such tasks, sensors
for these devices should be capable of detecting mechanical properties like strain, pressure and shear.
The mechanical properties of these sensors in terms of stretchability and Young’s modulus must match
the ones of human skin [6] if a direct interface to the body is desired. A benchmark for soft sensors is a
stretchability of up to 100% and even more, if joint angles are to be measured in wearable applications [7].

1.2. State of the art


In the past, many di↵erent sensing materials and designs were proposed in order to measure the aforemen-
tioned quantities at large strains. A good overview is provided by Zeng et al. [1]. Three sensing principles
are mainly used in the literature. Resistive sensing is based on a conductor that changes its geometry
when an external load is applied and is very popular for strain sensing [8–10]. Capacitive sensing is mainly
used in pressure or shear sensing and makes use of a compliant elastomeric sheet sandwiched between two
conductive plates [11, 12]. The third principle is based on carbon nanotube fibers and exploits piezore-
sistive properties when strained [13]. A key challenge for all sensor designs is to have a material that
remains conductive even at large deformations. Candidates are liquid metals such as eutectic gallium in-
dium [10], elastomers loaded with conductive particles such as carbon [14], silver [15], conductive polymers
blends [16], buckled graphene [17] and or carbon nanotubes [18] and ionic conductors [19]. All composite
approaches rely on a percolated network of conducting particles [15]. These devices usually exhibit a hys-
teresis due to the changing particle network under strain and cycle number [18,20], a phenoma called the
Payne e↵ect [21, 22]. Since the conductive network changes under strain and cycle number such devices
will always experience a drift in their resistance, which means that no reliable resistive sensor can be built
from such materials. Liquid metal sensors based on eutectic Gallium-Indium (eGaIn) do not display this
hysteresis, and various sensors for strain [23], pressure [10, 24], shear [9, 25] and even curvature [26] have
been developed. The material can be laser patterned and the fabrication of sensor arrays is possible [27].
Another sensor class that does not exhibit hysteresis is ionic strain sensors [28]. Ionic strain sensors based
on sodium chloride solution were first demonstrated by [29]. Chossat et al. [30] combined an ionic sensor
with eGaIn demonstrating extensible connections to the ionic sensor. Furthermore, the ionic solution can
be incorporated into a hydrogel network to yield a firm sensor material [28]. All ionic sensors require an
AC signal for readout, otherwise the electrodes degrade over time. Their time stability is also limited
due to rapid evaporation of the salt solution through the elastomer [28, 29]. Recently, this evaporation

15
1. Introduction

problem was solved by using a glycerol based sodium chloride solution [31]. All sensors also need an
encapsulation, support or in the case of capacitive sensing, a dielectric material. PDMS (Polydimethyl-
silane) [10, 12] is the most reported material for this task in literature. Other rubbery materials such
as SBS (styrene-block-butadiene-block-styrene) [15] and NBR (nitrile butadiene rubber) [32] have also
been used. Sensors come in various form factors such as fibers [33], fiber mats [33] or sheets [12, 27]. The
technologies used in order to fabricate these architectures include lithography [34], planar printing [35,36],
embedded 3D printing [20], micromolding [9, 30], multi-core shell extrusion [37], electrospinning [38, 39]
and dip coating [15,33]. For textile intergration, methodologies like sewing/attaching onto [7,16], directly
sewing into [40] and weaving into [41] fabric have been demonstrated.

1.3. Aim
We have clearly identified that there is a need for reliable soft strain sensors. In this thesis we explore
new soft sensor designs, which are specifically adapted for integration into a soft exosuit that is currently
developed [42]. There are several sensory feedback problems in the suit that need to be addressed in
the future. We will shortly state these and select the most promising solution by means of a design
process. Under the constraints identified, the concept of extruded strain sensitive fibers seemed most
feasible . Next, we will discuss the sensor design and propose an analytical model for it. We will then
present the development of printable materials and subsequently the nozzle design in order to extrude
the sensors. In the sixth chapter, we outline the extrusion process and elucidate the limits imposed by
the rheology of the printing inks and the nozzle design. Next, we will highlight how to transform the
extruded fibers into functioning sensors. Chapter 8 describes the readout circuitry and the software for
readout. In chapter 9, we charaterize the sensor’s static and dynamic response and compare it to the
proposed model. Finally, we demonstrate the applicability of the sensors by integrating them into textiles
and end with a conclusion and opportunities for further investigations.

16
2. Project definition
2.1. Problem statement and design goals
In this chapter we identify further research needs in soft sensing for the soft exosuit recently developed
in our lab. A brief overview of the soft exosuit can be found here [42]. Although a so called fancy
pants suit, which uses externally attached eGaIn sensor in order to measure joint kinematics has been
developed recently [7], a fully textile integrated solution of any kind of sensing is desired. The design
goal of this masterthesis is therefore to design and produce fully integrated sensor solutions that enhance
suit performance. In the next subsection, we explore the functional requirements of the sensing problem
on our sensor design.

2.2. Specifications sheets


The specification sheet defines the areas of research from the side of the customer, in this case the research
group.

2.2.1. Functional requirements


There are three main functional sensing areas of the suit identified that need further research and engi-
neering.

Joint Kinematics For feedback control of the suit real time gait cycle monitoring is required. Gate cycle
can be analyzed by measuring joint angles and velocities. External strain sensors that measure the change
in length between two points on the body surface can monitor these parameters [7] (Fig. 2.1). Values of
the most important lower limb joints were calculated and reported in [7]. A suitable sensor with a length
of 30 mm must be able to detect strains ranging from 130-150 % at the ankle joint, over 270-380% for the
hip, and up to 290-480% at the knee joint. Furthermore, for accurate control, the sensor’s tolerance must
not be more than ± 5 deg, otherwise forces are applied at the wrong portion of the gait cycle. It is also
important that the dynamic profile of the joint angle is detected accurately. Angular joint speeds range
from 140 deg/s (hip) over 160 deg/s (ankle) to 260 deg/s (knee) for a subject walking at a speed of 1.25
m/s [43]. In order to resolve these features at the required resolution, the sampling frequency should be
at least 110 Hz 1 . For an accuracy of ± 1 deg, even 550 Hz are required. The requirement on strechability
can be decreased, if larger sensors are used. Furthermore, for a 30 mm sensor, the sensing material must
be able to cope with a strain velocity in the order of 400-600 %/s while minimizing nonlinear e↵ects.

1 260deg/s = 52 samples/s ) 2 · 52 ⇡ 110Hz Whereas in the last step the nyquist criterion was used.
5deg

17
2. Project definition

Figure 2.1.: Relation between surface strain and joint angle in human body movements. Reproduced
from [7]

Tension and strain in the suit The stress distribution within the suit is required for optimizing the
force transmission to the joints and therefore the efficiency of the suit. The suit transmits approximately
200 N to the ankle. This force, transmitted by a bolden cable, is currently measured by a commercially
available loadcell (Fig. 2.2). However, this load cell is bulky. A flexible, compliant measurement solution
would improve device form factor and ergonomics. Currently, no method exists to measure the secondary
forces inside the nylon webbing of the suit. To assess webbing force distribution, a sensor requires a strain
measurement resolution below 1 %.

Figure 2.2.: Force transmission element to the ankle. Left: without loadcell, Right: with loadcell

Pressure on the skin To determine the transmitted mechanical moment from the suit onto the thigh,
the pressure exerted on the skin must be known. Knowledge of the pressure field would help to optimize
force transmission to the human body. No measurements on the pressure field currently exist. We
estimate that pressures on the thigh and shank are below 5 kPa, since this is the maximum normal force
for comfort [44]. The pressures on the iliac crest are expected to be much higher in the order of 300-500
kPa. A sensor array could sense this pressure field. The spatial resolution of the sensor array should be
in the order of 0.5 cm2 and the patch size is expected to be around 3 x 3 cm. The sensor should also
maintain its pressure sensing performance for strains up to 50 %.

2.3. Functional Specifications sheet


The functional specification sheet also includes all the constraints that act on the developer (the writer
of this thesis) of the new product. We want to evaluate which of the above research areas is the most
feasible in the scope of this masterthesis. For this reason, we list evaluation criteria, that we will use
in order to identify which area of research is most promising. We can group the criteria into six main
subcategories:

18
2.3. Functional Specifications sheet

• Temporal requirements: The timeframe granted to this project.


• Sensor requirements: The requirements that the sensor design must meet.
• Ergonomic requirements: All factors influencing the interface to the human being will be listed
here.
• Fabrication requirements: Requirements that deal with the actual sensor fabrication.
• Academic vs project requirements: All the issues of integrating this individual research project into
the overall project of the group and still make it a standalone thesis.
• Risk: All issues that could cause the project to fail or represent a major burden that needs to be
overcome are listed.
Figure 2.3 summarizes these criteria. This analysis is not exhaustive, but extending it would have been
outside the scope of this thesis. Requirements that were not considered in the analysis are for instance
financial, economic, scalability and legal requirements.

Figure 2.3.: Evaluation criteria for sensor solutions

2.3.1. Temporal requirements


Deadline A working prototype most be developed until the 30th of June. It is therefore of fundamental
importance that a project that is actually feasible in this timeframe is chosen.

2.3.2. Sensor requirements


An overview on the typical requirements of a sensor can be found in [45].

Robustness
• timestability (fatigue, evaporation, aging), o↵set drift, sensitivity drift
• low cross sensitivity, high selectivity (decoupling of pressure/strain, temperature e↵ect, strain speed
dependence)
• hysteresis, repeatability (materials nonlinearities)

19
2. Project definition

Sensitivity
• Resolution (smallest detectable increment)
• Threshold

Low Power The power consumption of a single sensor should be as low as possible. The suit is expected
to be in functional mode for 8 hours before recharging. For the whole sensor supply, no more than 10
Wh battery power is expected. The power consumption of a single sensor should be lower than 50 mA.

Textile integration Any sensor should be integrable into the fabric of the suit or into a textile that is
worn underneath the suit. Furthermore, the connections to the sensor should also be flexible and not
interfere with the normal function of the suit.

Easy Readout The readout electronics of the sensor should be as simple as possible.

Temporal Resolution The temporal resolution of the sensor should be at least 120 Hz, better 600 Hz.

2.3.3. Ergonomic requirements


Comfortable The sensor should be comfortable to wear and not cause any elevated pressure spots on
the skin.

Conformality The sensor should be elastic and stretchable to facilitate the natural movement of the
skin. For this the Young’s modulus of the sensing material needs to be below the Young’s modulus of
human skin, which is 15-150 MPa.

Machine washable Since the suit is worn by humans and intended for military applications any part
that cannot be detached must be machine washable.

2.3.4. Fabrication requirements


Easy Fabrication It should be easy to fabricate a first prototype in order to quickly gain some informa-
tion on the sensor specifications.

Mass Fabrication The sensor design should be scalable such that many sensors could be fabricated in a
short amount of time. For this reason, large quantity techniques such as spin-coating or extrusion should
be used for fabrication.

Available Technologies The possibilities largely depend on the available technologies. Other groups
at Harvard University have developed 3D manufacturing technologies which could prove helpful for this
project.

2.3.5. Academic vs project requirements


Most useful to project The research should be directly applicable to the soft exosuit.

Most scientifically interesting The research should be scientifically interesting, since a research paper
has to be written at the end of the thesis.

2.3.6. Risks
Evaporation This risks is associated with the sensing material. If the sensing material is a water
containing liquid, the water may evaporate over time and render the sensor non-functioning.

20
2.3. Functional Specifications sheet

S
Low Conductivity If the conductivity of the sensing material is below 10 3 cm the resistance of the
sensor might likely be in the M ⌦ range, which will render the readout more complicated.

Cross Sensitivity The sensor must be capable of discerning between pressure and strain stimuli.

Material Compatibilities The sensor will be composed of di↵erent materials. Each material should
interact with the other constituents in a non-detrimental fashion.

Complicated Readout Circuitry Some sensing principles will require complicated readout electronics.
Sensor arrays will require multiplexing and digital signal processing.

Soft/Hard Interface The readout electronics constists of metals and semiconductors therefore exhibit
a large Young’s modulus (GPa), whereas the sensor exhibits a very low Young’s modulus (kPa-MPa).
Such interfaces often result in mechanical failure of soft devices. This change in mechanical properties
cannot occur at one step, but must be gradually applied.

2.3.7. Relative importance of the functional specifications for later evaluation of


the developed solutions
In this section we developed a grading scheme, which reflects the importance of the di↵erent criteria.
The scheme is presented in Tab. 2.1.

Requirements PW OW
Temporal requirements 0.2
Deadline and overall feasibility 1.0 0.2
Sensor requirements 0.2
Robustness 0.2 0.04
Sensitivity 0.2 0.04
Low Power 0.1 0.02
Textile Integration 0.2 0.04
Easy Readout 0.1 0.02
Temporal Resolution 0.2 0.04
Ergonomic requirements 0.1
Comfortable 0.5 0.05
Conformity 0.4 0.04
Machine washable 0.1 0.01
Fabrication requirements 0.2
Easy Fabrication 0.5 0.1
Mass Fabrication 0.2 0.04
Current Technologies 0.3 0.06
Academic vs project requirements 0.2
Most useful to project 0.5 0.1
Most scientifically interesting 0.5 0.1
Risks 0.1
Evaporation problem 0.2 0.02
Low Conductivity 0.2 0.02
Cross sensitivity 0.2 0.02
Material compatibilities 0.1 0.01
Complicated readout circuitry 0.1 0.01
Soft/Hard interfaces 0.2 0.02

Table 2.1.: Weighting for the di↵erent requirements. OW marks the overall weight for this category, PW
the partial weight for this category.

21
2. Project definition

2.4. Part functions


Before we can develop a design, that can be evaluated according to the above mentioned criteria, we need
to split the design problem into subtasks, called part functions. For these part functions, we then search
solutions and based on them we will develop two design concept. From these we will subsequently select
the optimum sensor design. The sensor design can be subdivided into seven part functions:
• Sensing Principle
• Conductive Material
• Dielectric/Insulating/Piezoactive/Encapsulation material
• Sensor Architecture
• Sensor Fabrication
• Readout Electronics
• Sensor Integration

2.4.1. Solutions to part functions


The possible solutions were discussed in the introduction of this thesis and are shortly summarized here
in table 2.2.

Sensing Principle Conductive Material Encapsulation Material Sensor Architecture


Capacitive Liquid metals PDMS Sheets
Resistive Particle loaded elastomers NBR Fiber mats
Piezoresistive Ionic conductors SBS Fibers

Sensor Fabrication Sensor Integration Readout


electronics
Lithography Sewing onto fabric Resistive:
Wheatstone bridge
Planar Printing Sewing into fabric Capacitive: Either
a commercial
solution or a
custom built AC
measurement
technique
Embedded 3D printing Weaving into fabric
Micromolding
Multi-core shell extrusion
Electrospinning
Dip coating

Table 2.2.: Solutions to part functions

2.5. Concepts
Based on the possible solutions for the part functions, we developed two concepts, one for joint kinematics
monitoring and one for pressure sensing on the skin. For the tension in the suit no detailed concept was
developed, because a soft sensing approach did not seem to be appropriate for the small strain values of
the nylon fabric or for the load cell replacement at the bolden cable. The two developed concepts are
shortly described below.

22
2.6. Evaluation of the developed concepts

2.5.1. Concept 1 - Strain sensitive fibers for joint kinematics


The first concept uses extruded strain sensitive filaments that are directly woven into the suit for joint
angle monitoring (Fig. 2.4). The sensing principle could either be resistive or capacitive. To our
knowledge no capacitive strain sensing fiber has been mentioned in the literature. To accommodate
capacitive sensing, an extrusion setup and appropriate printing materials must be developed. Because
of the problems mentioned in the introduction with composite conductors, we chose to use either an
ionic liquid or a liquid metal approach for the conductive material. The encapsulation material was
chosen to be silicone (PDMS), since it is available commercially in formulations that can be printed and
subsequently cured (Smooth-on Inc.). For the readout electronics, commercial solutions for capacitance
determination should be evaluated first.

Silicone elastomer

Ionic conductor

Figure 2.4.: Concept 1 - four core shell capacitive strain sensor with extruder nozzle.

2.5.2. Concept 2 - Pressure and strain sensitive patches


This concept consists of a patch that is strain and pressure sensitive and can be woven directly into
the fabric (yellow filaments) (Fig. 2.5). The two perpendicularly arranged grids serve as a resistive
strain sensor, whereas the grid points act as capacitive pressure sensors. The patches could be fabricated
by a microfabrication approach, including lithography techniques, spincoating and micromolding. The
dielectric and encapsulation material could either be PDMS or THF (tetrahydrofurane) solutions of NBR
or SBS rubber. The conductive material for the channels can either be an ionic liquid or a liquid metal
conductor. While the material and fabrication techniques for this motif are comparable to the fiber style
sensor, the wiring and readout electronics will be much more complicated for patch type sensors.

fabric fibers

pressure sensitive patches


strain sensitive lines (capacitive)
(resistive)

Figure 2.5.: Concept 2 - Pressure and strain sensitive patch

2.6. Evaluation of the developed concepts


For every criteria listed in Tab. 2.1 the two concepts were ranked according to how well they were
expected to perform. A one was given when the concept was superior to the other. In this case a two was

23
2. Project definition

given to the inferior concept. If both concepts were considered to be equal a value of 1.5 was assigned
to both of them. Therefore the concept with the lowest sum is the most promising. The results of the
weighted rank numbers is shown in table 2.3. We will give a short reasoning for the ranking:

• Deadline and overall feasibility: Since the fiber concept is simpler we considered it to be more
feasible.
• Robustness: The fiber concept is only one part and does not have multiple connections.
• Sensitivity: The distributed pressure sensing concept can sense strain and pressure, whereas strain
is detected resistive. The resistance change [M ⌦] is expected to be much larger than a capacitive
change [pF ], therefore the sensitivity of the patch is likely to be superior.
• Low Power: The fiber concept is only one fiber and does not involve multiplexing.

• Textile Integration: The fiber is easier since it can be directly woven into the textile like a filament.
• Easy Readout: The fiber concept is superior, since it involves no multiplexing.

• Temporal Resolution: The fiber, since computing power is not multiplexed over several sensors.
• Easy Fabrication: The fiber, since sensor body can be printed in one run. For distributed pressure
sensing already the sensor body has multiple fabrication steps.
• Mass Fabrication: Both are equally scalable.

• Current Technologies: The pressure patch is easier, since printing setup for strain fibers will have
to be developed from scratch.
• Most Useful To Project: The fiber because it allows for feedback control of the suit.

• Most Scientifically Interesting: The pressure patch since in addition to strain sensing also pressure
sensing is possible.
• Evaporation Problem: Both are equally a↵ected.

• Low Conductivity: The channels of the pressure patch are likely to be smaller in diameter, therefore
their resistance is higher.
• Cross Sensitivity: The pressure patch will have the decoupling built in, therefore we expect it to
be less cross sensitive.

• Material Compatibilities: Since both are using the same or similar materials, the issues encountered
here, will likely be in the same range.

• Complicated Readout Circuitry: The fiber is less risky, since it consists of only one sensor.
• Soft/Hard Interface: The pressure patch is worse, since it consists of many channels that need to
be connected to rigid cables.

24
2.6. Evaluation of the developed concepts

Distributed
Distributed strain fiber
strain fiber pressure
Requirements pressure concept
concept sensing
sensing weighted
weighted
Temporal requirements
Deadline and overall feasibility 1 2 0.2 0.4
Sensor requirements
Robustness 1 2 0.04 0.08
Sensitivity 1 2 0.04 0.08
Low Power 2 1 0.04 0.02
Textile Integration 1 2 0.04 0.08
Easy Readout 1 2 0.02 0.04
Temporal resolution 1 2 0.04 0.08
Ergonomic requirements
Comfortable 1 2 0.05 0.1
Conformity 1 2 0.04 0.08
Machine Washable 1 2 0.01 0.02
Fabrication requirements
easy to fabricate 1 2 0.1 0.2
mass fabrication possible 1.5 1.5 0.06 0.06
current technologies 2 1 0.12 0.06
Academic vs project requirements
most useful to project 1 2 0.1 0.2
most scientifically interesting 2 1 0.2 0.1
Risks
Evaporation problem 1.5 1.5 0.03 0.03
Low Conductivity 1 2 0.02 0.04
Cross sensitivity 2 1 0.04 0.02
Material compatibilities 1.5 1.5 0.015 0.015
Complicated readout circuitry 1 2 0.01 0.02
Soft/Hard interfaces 1 2 0.02 0.04
Sum 25 38 1.235 1.765

Table 2.3.: The feasibility of the two concepts was investigated by ranking and weighting them according
to the criteria. A one was given when the concept was superior than the other. In this case a
two was given to the inferior concept. If both concepts were considered to be equal a value of
1.5 was assigned. The concept with the lowest weigthed sum is chosen, in this case the fiber
concept.

The lower weighted sum in column three in Tab. 2.3 suggest that the capacitive fiber concept is more
promising and should be chosen for the scope of this masterthesis. We will therefore discuss the necessary
steps and present the technology that was developed in order to make this concept a reality.

25
3. Sensor model
3.1. Introduction
We have identified capacitive strain sensitive filaments as the most promising sensor development task
for this thesis. In this chapter we derive a model for the capacitive strain sensor.

3.2. Model derivation


We will first describe the sensor analytically in its a strain free state and then see how the properties
change under strain.

3.2.1. Unelongated state


The sensor geometry can be modeled as a cylindrical capacitor with resistive electrodes. A conductive
core with a finite resistivity is surrounded by a dielectric, which in turn has another conductive layer
around it. This whole structure is then encapsulated by an outer stretchable layer as shown in Fig. 3.1.

r2 r3 r4 Conductive layers
r1
Dielectric layers

Figure 3.1.: Schematic highlighting the cross section of the sensor. The conductive layers are shown in
red, the dielectric layers in white.

The sensor behavior can be modeled via the equations for a cylindrical capacitor, a cylindrical resistor
and a cylindrical ring resistor. The capacitance of a cylindrical capacitor is given by:
2⇡✏✏0
C= ⇣ ⌘L (3.1)
ln rr21

where ✏ is the permittivity of the dielectric between the two capacitor plates. r1 , r2 are the radii as
depicted in Fig. 3.1 and L is the length of the sensor.
The resistance of a cylindrical structure is described by:
L L
R=⇢ =⇢ (3.2)
A ⇡r1 2
where ⇢ is the resistivity of the material, L is the length of the wire and A the crosssectional area.
The resistance of a cylindrical ring is given by:

27
3. Sensor model

L L
R=⇢ =⇢ (3.3)
A ⇡ (r32 r22 )
An infinitesimal slice of the sensor can be modeled by a series impedance consisting of two resistors and
a capacitor (Fig. 3.2).
Applying these equations on the infinitesimal element yields the following values for the resistance of an
infinitely small ring dRo , an infinitesimal cylindrical capacitor dC and an infinitesimal cylindrical resistor
dRi :
dL 2⇡✏✏0 dL
dRo = ⇢ 2 2)
dC = ⇣ ⌘ dL dRi = ⇢ (3.4)
⇡ (r3 r2 ln rr21 ⇡r1 2

dRo
dL
dC
dRo

dRi dC

dRi

Encapsulation
layer

Figure 3.2.: Cross section of the sensor with its equivalent circuit.

Connecting many of these slices in series would give rise to the distributed circuit model shown in the
left of figure 3.3.

....
dRo dRo Ro

dC dC dC Cs

dRi dRi Ri
....

Figure 3.3.: Equivalent circuit diagram of the sensor

To calculate the resulting impedance, between any of the four ports of this model, would require to solve
the partial di↵erential equations from transmission line theory [46]. However, the length of the sensor
will not exceed a few decimeters and the readout frequency is on the order of a few kHz, which renders
8
the wavelength to be = fc = 3·10 [m/s] 5
1000[Hz] ⇡ 10 [m]. Since there are six orders of magnitude di↵erence
between these two values, voltage and current amplitude can be regarded as constant along the length
of the sensor at a certain instant of time. This allows us to reduce the distributed circuit model to a
simpler lumped element model of the form shown on the right side of Fig. 3.3. The lumped element
consists of a capacitor between two resistors in series. On average, the electrons have to travel through
half of the sensor in order to arrive at their final location on the capacitor surface. Thus, the values for
the resistances for the inner and outer layer at zero strain result in:
L0 L0
2 2
Ro (" = 0) = ⇢ 2 2)
Ri (" = 0) = ⇢ (3.5)
⇡ (r30 r20 ⇡r10 2
and for the capacitance:

28
3.2. Model derivation

2⇡✏✏0
Cs (" = 0) = ⇣ ⌘ L0 (3.6)
ln rr20
10

3.2.2. Elongated state


When the sensor is elongated by a strain ", the length changes according to L = (" + 1) L0 1 . Under the
assumption of an incompressible material, which is adequate for rubbers and liquids [47], the triaxial strain
state must fulfill the volume conservation constraint "1 "2 "3 =1 . Taking into account radial symmetry
"2 = "3 = "r and setting "1 = " + 1 leads to the expression for the contraction in radial direction
1
"r = p"+1 . The radii as a function of strain are given by ri = pr"+1
i0
. Inserting this in the relationship
for the sensor capacitance results in:
2⇡✏✏
Cs (") = ✓ r200 ◆ (" + 1) L0 (3.7)
p
"+1
ln r
p 10
"+1

and after simplification:


2⇡✏✏0
Cs (") = ⇣ ⌘ (" + 1) L0 (3.8)
ln rr20
10

For the resistances:


2 2
(" + 1) L0 2 (" + 1) L0 2
Ro (") = ⇢ Ri (") = ⇢ (3.9)
⇡ (r30 2 r20 2 ) ⇡r10 2
whereas the total sensor resistance is the sum of these two:
2 L0 ✓ ◆
(" + 1) 2 1 1
Rs (") = Ri (") + Ro (") = ⇢ + 2 (3.10)
⇡ (r30 2 2
r20 ) r10
In summary, the sensor properties experience the following power law dependence on strain:

Cs (") / " Rs (") / "2 (3.11)


The sensor responds linearly and quadratically as a function of strain for capacitance and resistance,
respectively.
The sensitivities of the sensor are given by the derivative with respect to strain:
For the capacitance, we obtain:
✓ ◆
r20 @Cs (") 2⇡✏✏0
S ✏, , L0 = = ⇣ ⌘ L0 (3.12)
r10 @" ln rr20
10

Thus the sensitivity is a function of the geometry of the sensor, namly the original length and the ratio
of the original radii, and of the dielectric constant of the elastomer between the two conductors. For
the silicone dielectric used in this work, the dielectric constant was 2.3-2.8 [48]. Using this value, the
sensitivity is plotted as a function of the ratio of the two radii for di↵erent lengths in Figure 3.4. For
reasonable radii ratio from 1.05 to 1.50, sensitivity ranges between 25 - 250 pF per 100% strain.

1 The L L0
strain is defined by: " = L0

29
3. Sensor model

Figure 3.4.: Capacitive sensitivity change with respect to the ratio of the two radii of the dielectric.

For the resistivity we obtain a sensitivity of:


✓ ◆ ✓ ◆
r20 @Rs (") ⇢L0 1 1
S ✏, , L0 , " = = (" + 1) + 2 (3.13)
r10 @" ⇡ (r30 2 r20 2 ) r10
This expression varies as a function of strain.

3.3. Discussion
The sensor model shows that we can use di↵erent sensor properties for the strain measurements. In
the case of capacitance we even obtain a favourable linear transducer. The sensitivity is adjustable by
sensor design. Constraints on the achievable radii ratio will be imposed by the fabrication process and
discussed in the fabrication chapter. Furthermore, an in depth derivation on how to measure these sensor
properties will be given in Chapter 8. Next, we discuss material selection in terms of a material that is
conductive, stretchy and is in compliance with the 3D printing process envisioned.

30
4. Sensor Material development
4.1. Introduction
In order to facilitate printing inks must fulfill certain requirements, which are summarized with other
requirements in table 4.1. An overview is provided by Lewis et al. [49]. The ink should be shear thinning
(thixotropic), such that the sti↵ness of the ink drops inside the nozzle, where there is high shear stress and
increases as soon as the filaments exits the nozzle in order to keep the cylindrical shape of the filament on
the substrate. One way to obtain a shear thinning behavior is to have dispersed small particles (µm to
nm) that can form network in the solution under low shear stress. The conductive material should be non-
irritant to the human body and exhibit no hysteresis in electrical properties when stretched. Therefore,
no composite conductor should be used. The elastomer should have a sufficiently slow curing time that
allows it to be printed without curing inside the nozzle. In the cured state, it should exhibit a high
toughness, large extensibility (large strains) and a low Young’s modulus, such that not too large forces
are required to strain the sensor. Depending on the chosen material combination, further requirements
arise that we will discuss shortly after we have given an explanation for the chosen materials.

Conductive Ink Encapsulation


layer and
dielectric ink
conductive viscosity between
10-200 Pas
viscosity between 10-200 Pas shear thinning
shear thinning formstable
non-volatile (vapor pressure below 1 Pa) 1-2 hours potlife
no hysteresis upon stretch
non- irritant to human body
Cured State
High toughness
High extensibility
Low Young’s
modulus

Table 4.1.: Requirements for the two inks

4.2. Methodology
4.2.1. Ink preparation
For the conductive ink, we first tried to print eGaIn (liquid metal), but because of the high surface
tension, the printed filamentary structure was not stable, and spherical drops were formed (Fig. A.1
on page 95). As such, nofurther work was performed with eGaIn. Auqueous ionic conductors, namely
water sodium chloride solutions, were also evaluated. However, due to the high water permeability of
silicone, 1 these solutions evaporated quickly. Room temperature ionic liquids [50] were explored, but non-
published results from another Harvard Group (Phillip Röthemund, George Whitesides Group) suggested
than many of these materials are readily absorbed by silicone. In addition, the biocompatibility of room

1 http://www.elveflow.com/microfluidic-reviews-and-tutorials/the-poly-di-methyl-siloxane-pdms-and-microfluidic,

27.7.2014

31
4. Sensor Material development

temperature organic liquids is limited. We then explored non-volatile solvents for sodium chloride. A
concise collection of solvents with very low vapor pressure can be found in [51]. Of these glycerol was
found to exhibit the highest solubility for sodium chloride at room temperature (8.1 g for 95 % Glycerol
2
). Other salts that are very soluble in water like potassium acetate and sodiumbicarbonate were also
tested in glycerol, but despite a higher solubility they did not perform better in terms of conductivity.
An advantage of a glycerol salt solution is the biocompatibility of the formulation [31]. We therefore
decided to use this formulation as the base material and search for a rheology modifier that rendered
the solution thixotropic. We tested the following chemicals: Xantham Gum (Aldrich), Laponite (BYK
Additives and Instruments, Inc), Gelatin from porcine skin (Aldrich) and two PEG’s (molecular weight
1500 and 8 million). The experimental protocol was as very simple: Between 20-30 g of glycerol salt
solution were put into a 50 ml container. Subsequently, a few grams of the rheology modifier under
investigation were added under continuous stirring and left overnight. The next day the viscosity of the
solution was assessed by tilting the container. The only compound that allowed to change the viscosity
such that the ink did not flow out of the container under tilting was PEG 1500. We found the final
composition that we described below, by altering the ratio of PEG to conductive glycerol solution and
pressing the ink out of a 3 ml BD plastic syringe. The optimization criterion was that the pressure needed
to extrude the filament should be as low as possible, whereas the filament should still remain its shape
after extrusion. In terms of silicone we followed the same approach and developed two di↵erent silicone
inks (Smooth-on, Inc.), an Ecoflex® 00-30 with a final shore hardness 00 of 30 and a Dragon Skin® 10
ink with a final shore hardness A of 10. We now present the final formulations and then move onto their
rheological and electrical characterization. 3

4.2.2. Recipies
4.2.2.1. Ionic conductive Ink recipe
Polyethylene glycol with a molecular weight of 1500 (PEG 1500) (Aldrich), glycerol 99.5 % (Aldrich) and
sodium chloride (MACRON) were used as received. Sodium chloride was put in excess into glycerol and
left overnight at 100°C under continuous stirring on a hot plate. The saturated glycerol solution was
decanted from the excessive sodium chloride. This solution was then mixed with PEG 1500 in a ratio of
10:1, and under continuous stirring, heated to 100°C until all the PEG 1500 had dissolved. The solution
was degased for 5 min under vacuum and subsequently poured into 10ml BD plastic syringes and cooled
down to room temperature. During the cooling process a phase separation occured and the ink became
opaque and obtained the shear thinning properties required for printing.

4.2.2.2. Ecoflex Ink recipe


Ecoflex® 00-30 Part A and B, Thi-Vex® Silicone Thickener and Slo-Jo® Platinum Silicone Cure
Retarder were obtained from Smooth-On and used as received. For a usual formulation 60g of part A,
60g of part B, 1.2 g of Thi-Vex and 0.6 g of Slow-Jo were added in order to obtain the desired rheological
properties. Mixing and degasing was performed for 30s each using a planetary mixer (THINKY). This
mixture was then degased for an additional two minutes under vacuum. In order to transfer the ink into
a 10 ml syringe, without introducing any new air bubbles, the hole at the tip of a 60 ml syringe was
increased by using a DREMEL to approximately 1 cm in diameter and the ink was then gently withdrawn
into the syringe. Once in the 60 ml syringe, the ink could be easily injected into 10 ml BD syringes.

4.2.2.3. Dragonskin Ink recipe


Dragon Skin® 10 slow cure Part A and B, Thi-Vex® Silicone Thickener and Slo-Jo® Platinum Silicone
Cure Retarder were obtained from Smooth-On and used as received. For a usual formulation 60g of
part A, 60g of part B, 0.12 g of Thi-Vex and 0.6 g of Slow-Jo were added in order to obtain the desired
rheological properties. Before mixing, the solution was degased under vacuum for 2 min. Mixing and
degassing was performed for 30s each using a planetary mixer (THINKY). Transfering to a 5 ml BD

2 http://www.sy-kogyo.co.jp/english/sei/tech data/phase4.htm
3 All experimental data to the concerning the materials development process can be found in the folder: Materials devel-
opment Experiments

32
4.3. Results

plastic syringe was performed in the same way as for the Ecoflex. 5 ml syringes were used, because the
10 ml syringes broke when used in a syringe pump loaded with Dragonskin.

4.2.3. Rheological characterization


We wanted to assess the rheological properties of all developed inks in order to verify that they were indeed
shear thinning. For the conductive ink the electrical properties and the weight gain were monitored over
38 days in order to assess the non-volatility of the formulation. An additional rheological analysis was
performed after 30 days. We also characterized the rheological properties of conductive inks prepared in
di↵erently sized containers (10ml BD syringes and 50 ml plastic containers). Rheological characterization
was performed with a TA Instruments DHR3 rheometer. The inks for printing were characterized using
a 40mm cone and plate with an angle of 2° and a truncation gap of 550 microns. For the conductive ink
after 30 days, a 60 mm cone and plate geometry was used. The angluar frequency for all experiments
was 6.28 rad/s and the oscillation strain was gradually increased from 10 4 103 %. We assessed the
rheology of ink cooled in a 50 ml container and in 10 ml BD syringes after one day of ink fabrication.

4.2.4. Time stability assessment


Since both glyerol and PEG 1500 are hygroscopic [52], the ink was expected to absorb humidity from the
air. This process will lower the viscosity of the ink and likely increase the mobility of the dissolved ions
and hence the conductivity. For this reason, 23 petri dishes of size 5.5 mm x 1.2 mm, were filled with
approximately 10g of ink each. The ink was prepared according to the recipe in section 4.2.2.1. 10 samples
were used for the weight measurements and 13 for the conductivity measurements. The containers were
left at ambient conditions for 38 days, and the properties were measured every 2 - 4 days.

4.2.4.1. Electrical characterization


For the electrical characterization a benchtop conductivity meter (SPER SCIENTIFIC) was used. The
ink was put into a 10ml syringe which was cut in two pieces in the middle, such that ink material needed
was minimized. Since the ink exhibited a high viscosity, the ink was left in the conductivity meter for 24
hours for equilibration and the value obtained at the end of the measurement was used (see appendix A
for the transients). Because the ink used for the measurement had to be disposed of, we could only take
one datapoint per measurement.

4.2.4.2. Weight gain over time


For every datapoint, 10 containers were measured and the mean and standard deviation of the relative
weigth gain calculated.

4.3. Results
4.3.1. Rheological properties for printing
Fig. 4.1a shows the real (storage) and imaginary part (loss) of the shear modulus as a function of shear
stress. As long as they are constant as a function of shear stress, the material is in the linear viscoelastic
regime, exhibiting a constant viscosity for a constant oscillation frequency !. The viscosity is defined
as [53]:

G00 ( )
⌘( ) = (4.1)
!
We measured the following viscosities in the linear regime ( = 0.2 Pa, ! = 6.28rad/s):
• Ecoflex 00-30: 137.17 Pa s

• Dragonskin 10: 161.84 Pa s


• Conductive ink: 65.52 Pa s

33
4. Sensor Material development

Above a certain shear stress amplitude the complex shear modulus starts to decrease in amplitude and
the particle network in the material starts to break down. The yield stress is reached when the phase
00
angle , defined as tan( ) = G
G0 , surpasses 45 deg. Which means that the material behavior is now closer
to a liquid than a solid. The phase angle of the three inks investigated is plotted in Fig. 4.1b and the
yield stress amounted to:
• Ecoflex 00-30: 20.73 Pa
• Dragonskin 10: 0.17 Pa
• Conductive ink: 21.48 Pa

After yielding the viscosities decreased to ( = 100 Pa, ! = 6.28rad/s):


• Ecoflex 00-30: 13.48 Pa s
• Dragonskin 10: 44.88 Pa s
• Conductive ink: 14.72 Pa s

(a) Real and imaginary part (G0 , G00 ) of the shear modulus (b) Phase angle [deg] as a function of oscillation stress.
as a function of oscillation stress.

Figure 4.1.: Rheological behavior of the inks used for printing.

Since the sti↵ness increase of the conductive ink was due to a precipitation/growth process the final
viscosity was dependent on the cooling velocity and therefore on the size of the container, we characterized
inks cooled down in two di↵erent containers. Fig. 4.2 reveales that this hypothesis was indeed true, but
the changes are minor enough that the ink was sufficiently viscous for the extruded sensors to retain
shape and still be shear thinning and printable for both cooling conditions.

34
4.3. Results

Figure 4.2.: Container dependence of the cooling process on the final rheological properties of the ink.
Data of ink cooled in two 10 ml syringes and one 50 ml container is shown.

4.3.2. Time stability assessment


Fig. 4.3 shows a petri dish of the ink before the time stability measurement and after 38 days. It can
be seen that the inks appearance changed from a nearly opague to a transparent solution. The viscosity
also decreased from 65.52 Pa s to 0.046 Pa s over the same time frame. The rheological data is displayed
in Fig. 4.5. Fig. 4.4a shows the weight and conductivity change of the ink over 38 days. As expected
weight and conductivity increased over time and fluctuated with the amount of humidity in the air. The
steep increase in conductivity at 15 days was due to the complete dissolving of the precipitate and hence
a dramatic change in the viscosity. After the precipitates were completely dissolved the weight and the
conductivity fluctuated with the change in relative humidity. Whereas it took approximately 20 days
for the ink in the containers to be in equilibrium with the humidity of the air, this process was much
faster when the ink in an actual sensor was considered. Here equilibrium was reached after 2-3 days after
printing (small Figure in Fig. 4.4a).

(a) Ink right after fabrication (b) Ink exposed to ambient condi-
tions after 38 days

Figure 4.3.: The ink before water absorption and after 38 days.

35
4. Sensor Material development

(a) Conductivity and relative weight increase of ink over (b) Relative humidity during the time stability over time
time. The small figure shows the relative weight in-
crease of five sensors that were printed 24 hours before
data acquisition started.

Figure 4.4.: Change of electrical properties and weight gain of the ink over time.

Figure 4.5.: Rheological behavior of ink before and after water absorption

4.4. Discussion
The experiments confirmed that all the inks are shear thinning and viscous enough to be formstable after
extrusion (comparable to the viscosity of Ketchup 50-70 Pa s 4 ). The conductive ink was non-volatile.
However, the conductivity changed as a function of temperature and humidity. The amount of water
absorption of glycerol as a function of humidity is given in [54]. Therefore the sensor, when operated
in a resistive motif, will have environmentally dependent performance. Calibration for enviornmental
changes can be attained via small resevoir filled with ink that is attached to the sensor and does not
stretch but is subsequent to the same environmental changes. The conductivity of the ionic conductor
with 2-3 mS/cm will lead to a very high resistance (k⌦ - M ⌦) for a sensor 5 . This will impose a major
challenge for the readout electronics since the capacitance of a cylindrical sensor will be in the order
of a few PicoFarrad. 6 The materials used for the inks are non-toxic and biocompatible. However,
4 http://www.research-equipment.com/viscosity%20chart.html, 19.6.2014
5A L 1 10 1
cylindrical tube of length 10 cm and a diameter of 320 microns will have a resistance of R = ⇢ A = 0.2 ⇡ (160·10 6 )2

6 · 106 ⌦
6 For a cylindrical capacitor with inner radius 160 microns and outer radius 260 microns, length of 10 cm and a permittivity

of 2.5 (PDMS). The capacitance is C = 2⇡✏✏ ⇣ 0⌘ L = 27.5 [pF]


r2
ln r
1

36
4.4. Discussion

when used inside the human body the ink will have to be adjusted, since the current ink will uptake
water through any water permeable membrane, until the osmotic pressures inside the sensor and in the
environment will be the same. A possible solution is to use simulated body with adjusted rheology for
printing. Although, this sensor is then not usable for wearable applications due to the volatility of the
ink. Furthermore, if a non-liquid ionic conductor is required, for instance to reduce the cross sensitivity
of the pressure, a hydrogel swelled with glycerol/sodium chloride solution would be a possible solution.
A highly stretchable hydrogel that could be adjusted was demonstrated by Sun et al. [55]. To conclude,
the ink has the desired properties for printing and is non-volatile. The next chapter will now deal with
the actual design of the four core/shell printhead.

37
5. Nozzle Design
5.1. Introduction
In the last two chapters we first proposed the sensing design based on a four core/shell capacitive fiber
and then developed and characterized the inks for printing. This chapter now deals with the development
of the actual printhead, a four core/shell nozzle. We used 3D printed parts for the support and sacrificial
metal tubing for the alignment of the inner tubes. The sacrifical metal tubing had exactly a wall thickness
corresponding to the spacing between core and shell. The sacrificial tubing was used to align the nozzle.
Once the nozzle was aligned and fixed, the sacrificial tubing was removed.

5.2. Methodology
The process of designing a new printhead was an iterative process. In total, four generation of nozzles
were built, but we will only talk about the final printhead. Top and bottom parts were 3D printed with
a Connex500 multimaterial 3D printer, the printing material used was VeroBlue RGD840. 1 . Tubes were
obtained from McMaster (Sizes in Tab. B.1) and cut to the corresponding length with a DREMEL cutting
tool. For the functional nozzles, the tube lengths of the core/shell layers (Tab. B.2) and intermediate
layers (Tab. B.3) can be found in appendix B.

1 The solidworks and STL files for all generations can be found in the folder Designs/Nozzles

39
5. Nozzle Design

5.2.1. Nozzle assembly process

a) d)

b) e)

c)
f)

Figure 5.1.: The nozzle assembly process. a) Parts for the assembly. Topleft: 3D printed support with
groves for the tubes and epoxy glue. Topright: 3D printed top with holes for the syringe
connectors and square grooves for the epoxy. Bottom: The four core shell/tubes and the three
longer intermediate/sacrifical tubes. b) All tubes are put into each other like an antenna. c)
The epoxy glue is applied to the grooves. d) The intermediate tubes are removed after curing
of the epoxy. e) Epoxy is applied to the top part for sealing. Here care should be taken not
to use too much epoxy. f) The final generation 4 nozzle after assembly.

Below we describe the assembly process of a nozzle of the last generation (Fig. 5.1). 2 First, the bottom
and top part were printed with the 3D printer and the tubes cut to the required length (Fig. 5.1a).
Subsequently, the tubes were put into each other to give rise to the antenna like structure show in Fig.
5.1b. Next, epoxy glue was filled into the groves of the bottom part by the help of a syringe (Fig. 5.1c).
After this, the antenna like structure was gently inserted and everything left alone for 30 minutes in
order to give the epoxy time to cure. Then the intermediate tubes were removed by tweezers (Fig. 5.1d).
Fig. 5.1e shows how the epoxy glue was applied to the top part and finally after another 30 minutes of
curing the nozzle (Fig. 5.1f) was ready to use. The connectors for the syringe pumps were commercially
available barbs (McMaster, Part number: 51525K284) that were cut with a DREMEL tool approximately
in the middle such that they fitted exactly into the 5 mm holes of the top part. Gluing them was not
necessarily required for sealing but it improved the mechanical stability.

2 Videos of the assembly process can be found in the folder Videos.

40
5.3. Results

5.3. Results
The final printhead is shown in Fig. 5.2. The bottom part had a support structure for all tubes and
grooves for the epoxy which avoided that the four compartments were blocked in the gluing process.
All the core-shell tubes were of equal lenghts (12 mm) and they did not reach to the tip of the nozzle.
This reduced the freestanding part of the inner tubes and improved centering considerably. The exact
geometry is highlighted by a cross section through the nozzle in Figure 5.3. As it will become apparant
in the next chapter, the centering of this nozzle was sufficient for sensor fabrication.

1 cm

Figure 5.2.: The developed print head: Right top: top part, Right middle: bottom part, Right bottom:
Printhead tip enlarged

41
5. Nozzle Design

Figure 5.3.: Cross section through the final nozzle. The inner tubes had as short as possible freestanding
parts, which improved the centricity of the nozzle. Since the Reynoldsnumber Re = v·d ⌫ is
very low (viscous inks, slow velocities and small diameters) the flow will be laminar and as
long as the surface energies of the inks are not too dissimilar the interface will be stable and
mixing will not occur. We will see that this was indeed the case in the next chapter.

5.4. Discussion and Outlook


In this chapter, we have developed a print head that is capable of extruding a cylindrical fiber with four
di↵erent cylindrical layer and a filament diameter of a 1-2 mm. As we will see in the next chapter the
alignment of the nozzle was still not perfect. Furthermore, the achievable nozzle diameters and ratio were
limited to the available tubing sizes. However, it was enough for sensor fabrication. One possibility to
improve the current nozzle design would be to use a pop-up MEMS process [56] for the centering of the
tubes and to use pulled glass tubing in order to increase the number of possible diameters.

42
6. Printing Process development
6.1. Introduction
In the last chapter, we have fabricated nozzles that could potentially print a four layered filament. In
this chapter, we will present the setup used for printing, summarize the print parameters and analyze
the controllability and accuracy of the process. The evaluation of the optimal printing parameters was
iterative and we will only discuss the meaningful experiments in this report. For this reason, we are not
going to discuss them here. The structure of the chapter is as follows: First, we are going to present the
controllability of the filament thickness with print speed and evaluate the centricity of the layers. In the
second part of the chapter, we will provide the parameters that were used to create functional sensors.

6.2. Speed tests


In this section we are going to assess the layer alignment and how we can control the thickness by varying
the print speed. Furthermore, the maximum possible print speed was determined. The radii ratio between
each layer can also be altered by changing the volumetric flow rates of the individual layers, but we did
not explore this in the scope of this thesis.

6.2.1. Methodology
The inks were fabricated according to the recipies mentioned in the chapter Sensor Material Development
in sections 4.2.2.1, 4.2.2.2 and 4.2.2.3. Four syringe pumps (two PHD Ultra (Harvard Apparatus - 703007)
and two PHD 22/2000 (Harvard Apparatus - 702001)) with controlled volume flow rates were used in
order to pump the inks through the nozzle (Fig. 6.1). The nozzle itself (number 12 (Table of fabricated
nozzles: B.2)) was attached to a 3-axis CNC controlled station, ABG 10000 (Aerotech Inc.). The print-
height of the nozzle was approximately one centimeter above the substrate. However, print quality was
insensitive to nozzle height, as long as the filament did not start to neck when hanging down. After
printing the nozzles were cleaned with hexane in order to avoid curing of the silicone elastomer inside it.
The filaments were left at room temperature for a day in order to allow them to cure. Optical micrographs
were acquired with a VHX 2000 (Keyence) microscope. We will now analyze the speed tests for Ecoflex
and subsequently focus on Dragonskin.

43
6. Printing Process development

Inner Conductor

CNC
Dielectric Layer controlled
stage

Outer Conductor

Encapsulation Layer

Print head
Syringe Pumps
Printed Sensor Filament
(a) Concept image showing the printing of a four (b) The actual printing setup that was used for the
core shell cylindrical capacitor fiber onto a glass printing of four layered filaments. In this setup,
plate. Red: Ionic conductor, White: Silicone two supplies were on the same syringe pump. How-
elastomer ever, beside an additional syringe pump the setup
of the experiments of this chapter was identical.

Figure 6.1.: Overview printing

6.2.2. Ecoflex 30 Speed test


In total, 19 filaments of a length of 18 centimeters (Fig. 6.2a) were printed at three di↵erent speeds of
the print head, filament 1-7 were printed at a velocity of 5 mm/s, 8-14 at 10 mm/s and 15-19 at 7.5
mm/s. The flow rates were held constant for all three experiments. Table 6.1 shows the flow rates of the
syringe pumps for the di↵erent experiments and the diameters of the filaments to be expected at these
conditions. We collected the diameter data for these filaments under a microscope but did not analyze it,
since no characterized sensors were fabricated with Ecoflex 30. The images with the measured diameters
can be found in folder “Fabricated Sensor Prototypes/2 Printed Sensors and speed tests ecoflex 25 4
2014/Thickness of extruded cables”. Figure 6.2b shows two filaments printed at two di↵erent speeds 5
mm/s and 10 mm/s. It can be seen that no necking of the filament occured even for printspeeds of 10
mm/s. Fig. 6.2c shows an artifically colored image for better visualization of the individual layers.

Layer Diameters Extrusion d [mm] d [mm] d [mm]


of Nozzle rates vph 5 vph 10 vph 7.5
[mm] [µl/s] mm/s mm/s mm/s
Core 0.254 0.628 0.4 0.282 0.326
Shell1 0.6858 1.884 0.8 0.565 0.653
Shell2 1.1684 3.141 1.2 0.848 0.979
Shell3 1.9558 10.053 2 1.411 1.632

Table 6.1.: The print parameters and the predicted diameters d of the four layers for di↵erent printhead
velocities vph . The diameter values always include the smaller filaments.

44
6.2. Speed tests

a) b)

(a) The printed Ecoflex filaments (b) Two Ecoflex filaments. a) was printed at a velocity of 5
mm/s and b) printed at a velocity of 10 mm/s.

500

(c) Artificially colored crosssection


of a filament printed at a veloc-
ity of 5 mm/s

Figure 6.2.: Macroscopic and microscopic images of the printed Ecoflex filaments.

6.2.3. Dragonskin 10 Speed test


Since we used the Dragonskin silicone elastomer for sensor fabrication, we analyzed the data from these
speed test more closely. We printed Dragonskin filaments at a speed of 2, 4, 5, 6 and 8 mm/s. We
took cross section images of the filaments in order to assess the thickness of the individual layers and the
dielectric in particular as a function of print speed. The flow rates used were again held constant for all
experiments and can be found together with the predicted filaments diameters in Tab. 6.2. Higher flow
rates for the Dragonskin than the ones used were not possible, because otherwise the plastic syringes
(BD, 5 ml) got destroyed.

Layer Diameters Extrusion d [mm] d [mm] d [mm] d [mm] d [mm]


of Nozzle rates vph 2 vph 4 vph 5 vph 6 vph 8
[mm] [µl/s] mm/s mm/s mm/s mm/s mm/s
Core 0.254 0.417 0.515 0.364 0.326 0.297 0.257
Shell1 0.6858 0.754 0.863 0.610 0.546 0.498 0.432
Shell2 1.1684 1.257 1.243 0.879 0.786 0.718 0.622
Shell3 1.9558 4.021 2.026 1.433 1.281 1.170 1.013

Table 6.2.: Printing parameters together with the predicted layer diameters d of the four layers for dif-
ferent printhead velocities vph

Fig. 6.3 shows the top down images and and the cross sections for the three di↵erent speed tests. Since
the ionic conductive ink was liquid it drained out of the cut fiber and was not visible anymore in the
cross section images. The cross section did not remain circular but was deformed during deposition onto
the flat substrate. However, the dielectric and the innermost layer were not a↵ected by this process.
Furthermore, the innermost layer was not perfectly centered in the second one, leading to an asymmetric
dielectric. We will see in chapter 9 that despite poor centricity functional sensors could be obtained.

45
6. Printing Process development

Fig. 6.4 reveals a necking of the filament for print speeds of 6 mm/s and above, marking the minimum
filament thickness at 1.3 mm (Tab. 6.2 print speed 5 mm/s) that can be achieved with this nozzle
by pulling the filament. If a smaller filament thickness is required a smaller nozzle must be built. We
measured the thickness of the resulting layers from top down images and from crosssection images. The
raw data is summarized in appendix C in Tab. C.1 and Tab. C.2. Fig. C.1 and Fig. C.2 shows how the
data was acquired. Always five top down images and five crosssections were analyzed and mean and 95
% confidence intervals were calculated. The amount of o↵-centering of the dielectric was determined by
two circles aligned to the two innermost layers and by measuring the distance between their centers (Fig.
C.1b). Fig. 6.5 displays the thickness of the core (Fig. 6.5a), shell 1 (Fig. 6.5b), the average dielectric
thickness (Fig. 6.5c) and the thickness of the whole fiber (Fig. 6.5d) as a function of printspeed. It can be
seen that the top down image always yielded a larger value than the prediction from the volume flow rates.
Two reasons might have caused this. First, the inner layers might have appeared enlarged because the
layers have di↵erent refractive indices and the shape of the filament acts as a magnifying glass. Second,
the outer layers were enlarged, because gravity forced the cross section to assume a D-shape. Conversely,
the crosssection images suggested a lower value than expected. We will give two possible reasons for this
observation. First, the measurement method with the two circles did not capture all the material in it.
And second, there might be some shrinkage of the elastomer during curing. Nevertheless, despite this
small o↵set, all the curves showed the same trend, suggesting that varying the print speed provides an
accurate measure in order to reduce the thickness of the dielectric and the whole filament. This will allow
for a larger degree of miniaturization and a smaller dielectric, which results in a higher capacitance and
a higher sensitivity of the sensor.

Figure 6.3.: Printed filaments as a function of print speed - Top down (a, b, c) and cross sections (d, e,
f) images of printed filaments for di↵erent print speeds (2 mm/s (a, d), 4 mm/s (b, e) and
5 mm/s (c, f) )

Figure 6.4.: Necking at a print speed of 6 mm/s, marking the maximum print velocity for Dragonskin at
these settings.

46
6.3. Printing filaments for sensor fabrication

(a) Thickness of the core as a function of print speed (b) Diameter of the first shell (Dragonskin 10) as a func-
tion of print speed

(c) Average thickness of the dielectric as a function of print (d) Overall filament thickness as a function of print speed.
speed

Figure 6.5.: Thickness of di↵erent layers as a function of print speed. Predicted (volume flow rates) vs
measured values (top down and cross section images).

vph 2 mm/s vph 4 mm/s vph 5 mm/s


Dielectric minimum 69.2±13.92 54±11.27 47.8±11.76
Dielectric maximum 192.4±19.73 151.2±8.79 122.6±11.48
O↵set [µm] 61.6 ±8.02 48.6±6.35 37.4±4.22

Table 6.3.: Thickness and o↵-centering achieved for the described printing parameters.

6.3. Printing filaments for sensor fabrication


6.3.1. Introduction
We chose Dragonskin as the sensor material, because it is much more though than Ecoflex (102 pli1 vs
38 pli) 2 . Unfortunately, printing Dragonskin was much more challenging. The potlife of the ink was
approximately 45 min (2-3 hours Ecoflex) and the shear thinning behavior was less pronounced (Fig.
4.1). However, the superior mechanical sensor properties will be worth this inconvinience.

1 pli = pounds per linear inch


2 http://www.smooth-on.com/Silicone-Rubber-an/c2 1115 1129/index.html and http://www.smooth-on.com/Platinum-
Cure-Sili/c1115 1130/index.html

47
6. Printing Process development

6.3.2. Methodology
All the filaments were printed by nozzle number 12 (same as in last section). In total, we printed 34
filaments. The print settings can be found in Tab. 6.4 The reported thicknesses are for a printspeed of
4 mm/s. The printing process for a sensor filament was as follows: First, an end of 30 mm was printed
at a speed of 2 mm/s. Subsequently, the sensor body was printed at a speed of 4 mm/s for a distance
according to the desired sensor length (25, 50, 75, 100 mm). And finally, the velocity was again reduced
to 2 mm/s for another 30 mm to farbicate the second end.

6.3.3. Results
Fig. 6.6a elicits the printing process of the filaments and Fig. 6.6b the printed sensor filaments on the
glass slide after curing of a day. Fig. 6.6 displays the transition from an end cap to the sensor body. The
data of the determined top down values for the radii is summarized in Tab. C.3.

Sensors 1-8 Sensors 9-20 Sensors 21-34


Layer Extrusion d [mm] Extrusion d [mm] Extrusion d [mm]
rates [µl/s] rates [µl/s] rates [µl/s]
Core 0.251 0.283 0.417 0.364 0.417 0.364
Shell1 0.754 0.567 0.754 0.610 0.917 0.651
Shell2 1.257 0.849 1.257 0.879 1.257 0.908
Shell3 4.021 1.414 4.021 1.433 4.021 1.451

Table 6.4.: Predicted diameters d of the four layers according to the extrusion rates and the velocity of
the printhead vph

48
6.4. Discussion

(a) The nozzle magnified during printing of filaments for the sen- (b) The printed sensor filaments on the glass
sors of di↵erent lengths. A video of this process “Printing Sensor slide
cropped.mp4” can be found in the folder videos.

500 μm

(c) Transition from an end cap to the sensor body.

Figure 6.6.: Printing process of the sensor filaments

6.4. Discussion
We have shown in this section, that by altering the print velocity, we can change the thickness of the
filament during the printing process. This can be used in order to print larger cross sections at the
end of the filament, which allows for easier insertion of the electrical connection wires and smaller cross
sections in the middle. The layers displayed a certain degree of misalignment (50-70 microns) and also
exhibit a D-shaped crosssection, but this was not critical for sensor fabrication. Yet, if the centering of
the dielectric can be improved, it would allow at least the extrusion of a 50 µm dielectric. With the
current dimensions of the innermost tube of approximately 350 µm this will give a ratio of the radii of
r2 400
r1 = 350 = 1.14. In addition, the sensitivity could be further increased by increasing the innermost
radius. For instance, rr21 = 850
800 = 1.06 yields already very large capacitances in the hundreds of pF (Fig.
3.4). In the next chapter, we are now going to electrically connect the filaments and also attach a robust
interface for textile integration.

49
7. Sensor Fabrication
7.1. Introduction
In the last chapter we have described the process of obtaining four core shell filaments in the form of
cylindrical capacitors. The goal of this chapter is to describe the electrical connection and the design of
a robust interface from the compliant sensor to the rigid readout electronics.

7.2. Methodology
As all the other design tasks of this thesis, also the interface fabrication was an iterative process. We
started o↵ with copper connections and no sensor end cap for fabric integration. Subsequently, replaced
the copper by silver wires in order to avoid oxidation of the wire over time and finally ended up with a
robust interface in order to attach the sensor to the fabric. The early two sensors shown in Fig. 7.1a,b
were fabricated from some of the Ecoflex filaments described in section 6.2.2. Fig. 7.1c shows the final
sensor design. We are now going to outline the final assembly process. As the reader will realize there is
still a lot of optimization that can be done here, especially in terms of miniaturizing the end caps.

a) c)

b)

Figure 7.1.: Evolution of the fabricated sensors: a) the first sensor with copper connections and no end
cap. b) second sensor with silver connectors. c) final design with holes for fabric attachements

7.3. Electrical connection


A schematic of the electrical connection process is shown in Fig. 7.2. Images of the real process are
displayed in Fig. 7.3. A video “Electrical Connection Movie.mp4” showing the assembly process in detail
can be found in the folder “Videos”. First, we cut the sensor with a razor blade in an approximate
distance of 1 cm from the end. Second, we pulled out the inner tube by the help of a tweezer and cut it
in a way that the result looked like in Fig. 7.3c. Third, the first silver wire (127 µm, Surepure Chemetals
Inc) was inserted into the inner tube. Subsequently, Dragonskin shore hardness A 30 was injected in
order to insulate the two conducting layers. The recipe for this mixture was the following: Part A 10g,
Part B 10g and 1 g of Silicone Thinner obtained from Smooth-on, Inc, were mixed in a THINKY shear
mixer for 30 seconds and degased for another 30 seconds. After degassing the elastomer was loaded into
a syringe (BD, 3 ml) and injected with a 25 gauge needle. We chose Dragonskin 30, since it has a shore A
hardness of 30, which was more than the sensor body. By doing this we created a gradient in mechanical

51
7. Sensor Fabrication

sti↵ness within the filament. In the last step, the second wire was inserted, connecting the outer layer.
The fiber was left to cure before steps b), c), and e) were applied to the other end of the fiber.

Figure 7.2.: Schematic illustrating the electrical connection process. a) printed sensor filament, b) end
cut with razor blade, c) cutting of the inner filament d) insertion of the first silver wire, e)
injection of Dragonskin 30 and f) insertion of the second connection wire.

Figure 7.3.: Microscopic images illustrates the electrical connection process. a) printed filament, b) end
cut with razor blade, c) cutting of the inner filament d) insertion of the first silver wire, e)
injection of Dragonskin 30 and f) insertion of the second connection wire.

7.4. End cap fabrication


After curing, an end cap for easy textile integration and wire security were fabricated. The end caps were
fabricated by a molding process. The molds were printed with a Object30 3D printer and the files can

52
7.4. End cap fabrication

be found in “Designs/End cap molds”. We used Dragonskin 30 as the elastomer. A harder elastomer
was used in order to enhance the gradient in mechanical sti↵ness to closer match it to the hard electrical
connection. The recipe was as follows: Part A and Part B were mixed in a 1:1 ratio (THINKY, planetary
mixer) for 30 sec and degassing was performed for another 30 sec. After the sensors were put into the
molds (Fig. 7.4a) the elastomer was poured in and degased under vacuum for 1 min. Subsequently, a
razor blade was used in order to remove excess elastomer from the top of the mold. After curing overnight
the sensors were removed from the mold (Fig. 7.4b). With this design of the electrical connection, the
silver wire could still be pulled out easily. To further enhance connection strength,, we used a needle
in order to form a loop of silver wire at the end of the end cap (Fig. 7.4c) which greatly enhanced the
resistance against wire pullout. Lastly, the outer part of the wire loop was covered with silicone glue
(Sil-Poxy, Smooth-on Inc.). Fig. 7.5 shows a schematic of a final sensor, with a focus on the electrical
connection setup.

a) b)

c) d)

Figure 7.4.: The end cap fabrication process: a) sensor filaments with electrical connections in a mold,
b) Sensor filament with end caps after demolding, c) Neddle in order to make a loop into the
wire for better connection strength, d) Application of silicone glue in order to insulate the
connection.

53
8. Readout
8.1. Introduction
The purpuse of this chapter is to develop a readout circuitry that is capable of measuring the sensor
capacitance and resistance as a function of strain and derive an analytical model of this entire system.
The readout design is especially challenging, namely because a very capacitance (pF ) is sandwiched
between two large resistors (k⌦ M ⌦). First, we explored a commercial capacitance to voltage converter
(Analog Devices, AD7747). However, this device can only operate up to parasitic series resistances of 10
k⌦ according to the data sheet and is therefore not suitable for this application. Second, a capacitive
touch sensor controller (Freescale semiconductor, MPR121) was considered. This device applies a constant
current for a defined time to the capacitance to be measured and determines the voltage drop over the
capacitance after charging. Since both, charge and voltage are known, in theory capacitance can be
calculated from C = VQ . Yet, this approach did not yield a physically reasonable capacitance value, since
the high output resistance led to a lower output current than the electronics assumed. Therefore, the
measured capacitance actually decreased with strain, since the sensor resistance increased. Thus the
charging current was at an even lower value than what the readout circuit expected. The same problem
occured with a common LCR meter (TENMA, 72-10465 LCR Meter). Fourth, we investigated the three
voltage measurement method [57], which consists of applying an AC signal to a circuit of an unknown
impedance (sensor) in series with a measurement resistance and measuring three RMS-voltages: The
voltage phasor US over the sensor, the phasor UM over the measurement resistor and the phasor U over
sensor and resistor. From these values, the real and imaginary part of the sensor impedance could be
calculated. Lastly, we investigated two approaches that measure the decay time of the voltage across
a resistor in series with the sensor when a step response was applied to the entire system. These two
approaches will be described in the next two sections.

8.1.1. Ideal readout circuit model


In this section, we derive an analytical model for the readout of the sensor capacitance and resistance.
As mentioned in the introduction, we can measure these quantities, by applying a step response to the
system and measuring the voltage across a measurement resistor in series with the actual sensor. An
equivalent circuit model highlighting this principle is shown in Fig. 8.1.

Ro Step
Cs
Ri
Ri
Cs

Vm ( t )
Ro
Rm
Encapsulation
layer

Figure 8.1.: Equivalent circuit of the measurement setup for the readout of the sensor properties. Rm is
the measurement resistor and Vm (t) the decaying voltage across it. Ro , Ri and Cs are the
sensor properties introduced in chapter 3.

By applying Kirchho↵’s loop rule to this circuit we find:

55
8. Readout

Ro · i (t) + Vc (t) + Ri · i (t) + Vm (t) = Vapp (t) (8.1)


using Vm (t) = Rm ·i(t), i(t) = Cs dV
rearranging and defining Rs = Ro +Ri as the total sensor resistance,
dt
C

we obtain the following two equations:

dVC (t) dVC (t)


Rs Cs + Rm C s + VC (t) = Vapp (t) (8.2)
dt dt
and

dVC (t)
Vm (t) = Rm Cs (8.3)
dt
The first-equation describes the time evolution of the voltage drop over the sensor capacitance as a
function of time and the applied voltage. The second equations provides a link between the measurement
voltage and the capacitance voltage. The first equation is a first-order-di↵erential equation of the form:

dy (t)
⌧· + y (t) = u (t) (8.4)
dt
The transfer function of this system is:
1 1
H (s) = = (8.5)
⌧s + 1 (Rs Cs + Rm Cs ) s + 1
The step response of this system is given by the following well-known equation for a first-order system 1 :
t
y (t) = y(1) + (y (0) y (1)) e ⌧ (8.6)
the decay time is given by:

⌧ = Rs C s + Rm C s (8.7)
Initially, no charge is on the capacitor and the voltage across it is zero. After a couple of decay times
the capacitor holds its maximum amount of charge and all the applied voltage decays over it. Inserting
the capacitor voltage at the beginning and end VC (0) = 0 and VC (1) = Vstep results in the general step
solution of this system:
t
⇣ t

Vc (t) = Vstep + (0 Vstep ) e ⌧ = Vstep 1 e ⌧ (8.8)

If we insert this expression into equation 8.3 for the measurement voltage, we can describe the measure-
ment voltage with only passive components and the magnitude of the step response:
Rm Cs t Rm t
Vm (t) = · Vstep · e ⌧ = · Vstep · e ⌧ (8.9)
⌧ Rm + Rs
Since Rm and Vstep are known and Vm (t) and t are measurable, the two parameters Rs and the decay
time ⌧ can be obtained via a least-squares model fitting.
If we define the value of the measurement voltage at t = 0 as:
Rm
Vm0 = Vm (t = 0) = Vstep (8.10)
Rm + Rs
we obtain:
t
Vm (t) = Vm0 · e ⌧ (8.11)
This is a common exponential decay with initial value Vm0 and decay time ⌧ . From the linearized
measured step response:
1
ln (Vm (t)) = · t + ln (Vm0 ) (8.12)

1 http://lpsa.swarthmore.edu/Transient/TransInputs/TransStep.html

56
8.1. Introduction

we can deduce the decay time ⌧ and the measured step response Vm0 at t = 0 from a linear least squares
fit y = a · x + b via:
1
⌧= Vm0 = eb (8.13)
a
The sensor quantities can then be calculated from equation 8.10, which is nothing but a voltage divider
at t = 0 and equation 8.7 for the decay time :
✓ ◆
Vstep
Rs = 1 Rm (8.14)
Vm0
⌧ ⌧
Cs = =⇣ ⌘ (8.15)
Rs + Rm Vstep
·R
Vm0 m

These equations describe how to obtain the sensor quantities in general and not specifically as a function
of strain. In the next section, we will briefly describe the second algorithm and then discuss how the
strained sensor quantities relate to the unstrained ones with strain as the only changing variable.

8.1.1.1. Alternative methodology using a multiplexer and two resistors


This method was very similar to the previous one, except that two measurement resistors were used
that can be connect indiviually to the circuit to act as the measurement resistor. An equivalent circuit
diagram is shown in Fig. 8.2. This methodology was used first but then abandoned, because of numerical
problems (extinction) of the algorithm. A very brief describtion shall be given here: A multiplexer
selects one of the resistors (1 M ⌦) and subsequently a step response is applied and acquired. Next, the
multiplexer switches the measurement resistor (470 k⌦) and a second step response is acquired. From
the two expression for the decay times:

⌧1 = (Rs + Rm1 ) (Cs + Cp )


⌧2 = (Rs + Rm2 ) (Cs + Cp )

where CP is the parasitic capacitance of the circuit, we can solve for the overall capacitance C = Cp + Cs
and sensor resistance Rs :
⌧2 ⌧1
C= (8.16)
Rm2 Rm1
⌧1 Rm2 Rm1 ⌧2
Rs = (8.17)
⌧2 ⌧1
Since the capacitance was very small, the two terms ⌧1 Rm2 , ⌧2 Rm1 were very similar. Therefore already
the noise was sufficient in order to lead to sign changes of the resistance as it was observed experimentally.
This algorithm was therefore not robust enough and even slower than the one described in the last
subsection using only one resistor and the time information of the signal. For this reason, we will use the
previously described readout method and consider now the strain depedence of the decay time.

57
8. Readout

Step

Ri

Cs
Rm1

Ro
Rm 2

Figure 8.2.: Equivalent circuit model for the readout using a multiplexer (switch) in order to connect two
di↵erent measurement resistors Rm1 , Rm2 to the sensor and to acquire two decay times.

8.1.2. Strain dependence of decay time


As we have seen in chapter 3 the change in sensor capacitance and sensor resistance under a strain " are
described by the following equations:
2⇡✏✏0
Cs (") = ⇣ ⌘ (" + 1) L0 (8.18)
ln rr20
10

2 L0 ✓ ◆
(" + 1) 2 1 1
Rs (") = Ri (") + Ro (") = ⇢ + 2 (8.19)
⇡ (r30 2 r20 2 ) r10
In order to calculate the strain dependence of the decay time, we have to use equation 8.7 and insert the
two expressions above. We obtain the following expression for the decay time:
✓ ◆
2 3 1 1 1 2⇡✏✏0
⌧ (") = ⇢✏✏0 L0 (" + 1) ⇣ ⌘ 2 2)
+ 2
+ Rm ⇣ ⌘ (" + 1) L0 (8.20)
r20
ln r10 (r 30 r 20 r 10 ln rr20
10

This gives the following power law dependence for the decay time:

⌧ (") / "3 (8.21)


As a short reminder the power law dependence for the capacitance and resistance were Cs (") / " and
R (") / "2 . So far, we have assumed an ideal equivalent circuit diagram. In reality, there are parasitic
capacitances of the readout circuitry that need to be included. In the next chapter we are going to
address this issue and derive equations that will later be used in order to deduce the sensor parameters
from experiments.

8.1.3. Non-ideal sensor model


In the non-ideal case, we have to take into account the parasitic capacitances of the readout circuitry.
Any parasitic resistance is negigible since the sensor resistance will be in the k⌦ M ⌦ range. The total
capacitance is:

C (") = Cs (") + Cp = Cs0 (" + 1) + Cp = Cs0 " + Cs0 + Cp (8.22)


2⇡✏✏
with Cs0 = ⇣ 0 ⌘ L0 .
r This expression contains two fitting parameters Cs0 , Cp that can be obtained from
ln r20
10
a linear fit to the capacitance change as a function of strain.
If we write the resistance in the same form:
2
Rs (") = (" + 1) Rs0 = ("2 + 2" + 1)Rs0 (8.23)

58
8.2. Implementation

L0
⇣ ⌘
with Rs0 = ⇢ ⇡2 (r30 2 1 r20 2 ) + r12 . Then this expression contains only one fitting parameter Rs0 , which
10
can be obtained by a quadratic fit to the resistance change as a function of strain. The measurement
resistance Rm has not to be taken into account, since the o↵set is accounted for by equation 8.14.
If we insert equations 8.23 and 8.22 into the expression for the decay time 8.7 we get

⌧ (") = (Rs (") + Rm ) (Cs (") + Cp ) (8.24)


⇣ ⌘
2
= (" + 1) Rs0 + Rm ((" + 1) Cs0 + Cp ) (8.25)

We can write this in a simpler form by collecting terms by powers of strain.

⌧ (") = Cs0 Rs0 "3 (8.26)


2
+ (Cp Rs0 + 3 Cs0 Rs0 ) "
+ (2 Cp Rs0 + Cs0 Rm + 3 Cs0 Rs0 ) "
+ Cp Rm + Cp Rs0 + Cs0 Rm + Cs0 Rs0 (8.27)

This fitting equation contains three unknown parameters, Rs0 ,Cs0 and Cp , which in theory can be ob-
tained by a cubic fit to the decay time data plotted as a function of strain. However, we will determine
Rs0 ,Cs0 and Cp from a linear and quadratic fit to the cpacitance and resistance data respectively and
use these expressions in order to validate the model for the decay time.
Having derived the analytical equations that relate the change in strain to physically measureable quan-
tities, we will now proceed in describing how to implement this model in terms of hardware and software.

8.2. Implementation
8.2.1. Hardware
The hardware implementation was done on a microcontroller that was as small as possible and still
capable of yielding fast enough and accurate sensor readings. Furthermore, the software development on
this hardware should be as simple as possible. For these reasons, the ATMEGA 328P microcontroller in
the form of an arduino pro mini (Arduino) was selected. The package dimensions are smaller than one
square centimeter and it comes with a built-in development environment. (Arduino). The chip runs at a
clock frequency of 16 MHz. The power supply and digital signals are 5V in amplitude. The connection
setup is shown in Fig. 8.3. The step input to the system was applied via a digital output (Pin 3). This
output was connected to the core ionic conductor of the sensor (red). The ionic conductive shell of the
sensor was then wired up to the measurement resistor (blue). The voltage over the sensor was acquired
by an analog input (Pin A0) of the microcontroller (yellow). The other end of the resistor was connected
to ground (black). The basic signal flow is depicted in Fig. 8.4. After acquisiton of the step response by
the ADC of the microcontroller (75kHz, 8bits), the signal was cropped and linearized. Subsequently, a
linear fit was performed, which then allowed for the determination of the decay time and thus capacitance
and resistance. These values were then transmitted via serial to a computer.

59
8. Readout

Figure 8.3.: Connection setup for the readout electronics.






Figure 8.4.: Signal flow in the readout electronics setup.

8.2.2. Software
The software consisted of a microcontroller part written in C++ and data acquisition and visualization
written in Python. The software was only meant for data aquisition during the project and a proper
documentation was not written for it. However, some code might be reused in further developments of
the algorithm and new implementations of the GUI for sensor testing. For this reason, we will shortly
describe the most important scripts and where they can be found on the accompanying CD. We will first
present the implementations on the Arduino and subsequently the ones in Python.

8.2.2.1. Implementations on arduino


In total, four arduino script for di↵erent functions were implemented:
• “1 Sensor functionality assessment.ino”
• “ 2 Static readout half signal for linear fit.ino”
• “3 Dynamic Readout half signal for linear fit dynamic measurements.ino”
• “4 Dynamic Readout half signal for linear fit dynamic measurements with synchronization chan-
nel.ino”

60
8.2. Implementation

They are all similar, and the files can be found in “Electronics and Software/Sensor Readout one resistor/
Arduino files”. The first script “1 Sensor functionality assessment.ino” was used in order to assess whether
the sensors work. The second script “2 Static readout half signal for linear fit.ino” was used for data
acquisition and visualization of the sensor parameters - decay time, resistance and capacitance - for static
testing. These values were conveyed over serial continuously, if the start command from Python was sent
to it. The last two files “3 Dynamic Readout half signal for linear fit dynamic measurements.ino” and “4
Dynamic Readout half signal for linear fit dynamic measurements with synchronization channel.ino” were
used for dynamic testing, such as walking monitoring and dynamic testing as described in the following
chapters (9, 10). In addition to the sensor properties, they both convey a time signal and the second
script a reading of another digital input, that can be used to synchronize the arduino with other data
acquisition equipment such as the Instron Mechanical Testing machine. After this short overview, we will
now explain the file “2 Static readout half signal for linear fit.ino” in detail.

#include <linreg.h>
#include <math.h>

// #define DECAYMONITORING;
#define CAPACITANCE MONITORING;
// #define LOOP SPEED MONITORING;

First, the libraries required are included. All arduino files need the a linear regression library (lin-
reg.h/linreg.cpp for proper operation). Then three preprocessor directives can be set, which change what
the script will output over serial. These are:
• DECAYMONITORING - Here the decaying signal, the linearization and the linear fitting coeffi-
cients are transmitted via serial.
• CAPACITANCE MONITORING - Here the sensor quantities in addition to the loop execution
time are transmitted via serial.
• LOOP SPEED MONITORING - Here only the loop execution time is transmitted via serial.
To get started, use a serial monitor, hook up the readout electronics and see what each option outputs
over serial.

const int Signal = 3; // the number of the LED pin


int Sensor outputR1 = A0;

float R1 = 1.000000; // Resistance is in MOhm float


float Rs = 0.0; // Sensor Resistance
float Cs = 0.0; // Sensor Capacitance

// #
// variables for the data aquisition
const int buffersize = 50;
int TimeSignalR1[buffersize];
int SensorResponseR1[buffersize];

// #
// variables for the data processing

double LogarithmOfSignal1[buffersize];
double TimeSignalhilf[buffersize];
double a1;
double b1;
double Rsquare;
int IndexDecayStart1;
int IndexDecayEnd1;
bool SendDataPoint = true;

// #
// variables for averaging

61
8. Readout

double sum tau = 0.0;


double sum intersects = 0.0;
double sum weights = 0.0;
double average tau = 0.0;
double average intersects = 0.0;
int average size = 10;

Second, all global variables are defined. The first section contains the variables for the external connections
of the arduino and the value of the sensor resistance. Subsequently, all variables for the acquisition of the
decaying voltage signal are defined, the size of the bu↵er can be adjusted to the decay length. Next, the
variables for the linearization and the cropping of the data are defined. Lastly, one can set an average
bin size that prefilters the signal before it is transmitted via serial. This is not a moving average filter,
it takes a certain number of samples, averages them and empties it completely for the next set.

void setup() {
// set the digital pin as output:
pinMode(Signal, OUTPUT);

// Increase the analog sample rate to 75 kHz.


ADCSRA &= ¬( BV(ADPS0) | BV(ADPS1) ); // clear bits 0 and 1.
ADCSRA | = BV(ADPS2); // set bit 2.
// analogReference(INTERNAL);

Serial.begin(115200);
}

Third, the setup function sets the pinmode of the pin, where the signal is applied to, to output. The
next lines change the standard settings of the ADC (10 kHz and 10 bit accuracy), because they are too
slow in order to have enough data points to fit the decaying signal. At the cost of two bits accuracy, the
sampling rate is increased from 10 kHz to 75 kHz. The last line launches the serial communction at a
baudrate of 115200.

void loop() {
#ifdef LOOP SPEED MONITORING;
unsigned long loopStartTime = micros();

#endif

#ifdef CAPACITANCE MONITORING


unsigned long loopStartTime = micros();
#endif

digitalWrite(Signal, LOW);
sum tau = 0.0;
sum weights = 0.0;
average tau = 0.0;
sum intersects = 0.0 ;
average intersects = 0.0;

// the signal swichting ability has to be implemented as well.

for(int i=0; i < average size; i++){

Measure Time response(Signal,Sensor outputR1,TimeSignalR1,SensorResponseR1, buffersize);


AddDataPoint(calcDecayTime1());

#ifdef DECAYMONITORING
DoDataLogging("RawSignalR1",TimeSignalR1,SensorResponseR1,buffersize,1);
DoDataLogging("LogarithmR1",TimeSignalhilf,LogarithmOfSignal1,IndexDecayEnd1 IndexDecayStart1,4);
DoDataLogging("Linfit1",&a1,&b1,1,7);
#endif

CalculateAverage();
calcCapacitance();

62
8.2. Implementation

calcResistance();

CheckforSerialMessage();

#ifdef CAPACITANCE MONITORING

if(SendDataPoint == true){
Serial.print(average tau,4);
Serial.print(",");
Serial.print(Cs,4);
Serial.print(",");
Serial.print(Rs,4);
Serial.print(",");
Serial.print(micros() loopStartTime);
Serial.print(",");
Serial.print("\n");
}

#endif

// Serial.println(micros() current millis);

#ifdef LOOP SPEED MONITORING;


Serial.println(micros() loopStartTime);
#endif
}

Then the main loop is called. According to which preprocessor directive is set, di↵erent versions of the
code are executed. First, all variables used for averaging are set to zero. Second, a for-loop that applies a
step response and acquires the decaying signal is executed . All functions mentioned from now on can be
found in appendix D.1. In the for-loop, the function AddDataPoint() calls the function calcDecayTime1()
which does the processing of the time signal, returns the decay time and sets the global variables of the fit.
The function AddDataPoint() then stores the weight for the averaging, the decay time and the intersection
at zero as a sum. CalculateAverage() calculates the average from these sums. calcCapapacitance() does
the capacitance calculation from the averaged quantities according to equation 8.15. calcResistance()
performs the resistance calculation according to equation 8.14. The function CheckforSerialMessage()
checks whether the computer has sent anything over serial and performs actions according to the messages
obtained.
A special comment is devoted to the function IndexDecayStartandEnd(), especially how this function
does the cropping for the linear fit. IndexDecayStart1 and IndexDecayEnd1 mark the beginning and end
of the cropping interval. First, the index of the maximum is calculated and IndexDecayStart1 is set two
samples beyond it. Next, the index, where the signal has dropped to 13.5 % of the maximum (two decay
times), is set as the IndexDecayEnd1. Then a bit less than half of this interval is abandoned by setting
the IndexDecayStart1 to half of the IndexDecayEnd1.
The signal is cropped in the middle as described in section 9.2. The script “1 Sensor funcitonality
assessment.ino” does not do this additional cropping in the middle. The script “3 Dynamic Readout
half signal for linear fit dynamic measurements.ino” and “4 Dynamic Readout half signal for linear fit
dynamic measurements with synchronization channel.ino” both do it. We also implemented the version
that used the multiplexer and two resistors for readout. A brief describtion of the files (Arduino and
Python) is given in appendix D.2, since they could serve as a starting point for the implementation of a
GUI for sensor calibration and storing the calibration values on the arduino.

8.2.2.2. Implementations in Python


In general, all python files contain external path definitions that need to be adjusted on every computer.
All the custom developed library files that are used by multiple scripts are contained in the folder: Elec-
tronics and Software/Python custom libraries. Furthermore, frequently used standard python libraries
include: scipy, numpy, pandas, matplotlib, pyqtgraph, time and pySerial.
Three GUI’s and three data acquisition scripts were written:
• 1 GUI Monitoring of decay.py - This GUI can be used with any of the four arduino files as long
as the preprocessor variable DECAYMONITORING is set. It shows up to two decaying voltage
signals their linearization and the linear fit through them. A screenshot of it is show in Fig. 8.5.

63
8. Readout

Important: This script freezes, when no appropriate data is send over the serial port. Furthermore,
the correct serial port has to be specified in the file Shared functions.py in the function connect
Serial().

• 1 Monitoring of decay Data acquisition.py - This script allows for data acquisition of the decaying
voltage signal, the linearized data and the coefficients of the linear fit. Everything is saved as a csv
file, whereas path, sensor number and amount of strain can be specified.

• 2 GUI Monitoring of Sensor quantities.py - This GUI is meant for visualization of the sensor
quantities. Not all its functions were implemented. Especially, data acquisition was done by simple
script files. The calibration was also not fully implemented. It is to be mentioned that the looprate
of the GUI is not too high and if there is no averaging performed of the values on the arduino it
usually freezed (an averaging number above 10 usually resolved this issue) (Fig. 8.6)

• 3 Data acquisition for Electronics Evaluation.py - This is the script that was used to save the
csv-files for the evaluation of the readout electronics.

• 4 Static Data acquisition Sensors.py - This script was used for the data acquisition of the static
sensor testing. It just writes every received line over serial into a csv file. The number of lines to
be written can be specified in the script.
• 5 Dynamic Data acquisition Sensors.py - This script was used for dynamic data acquisition. It does
beside having a di↵erent filename the same as the previous script.
• 6 GUI one large screen for resistance monitoring.py - This is a GUI for demonstrations of the sensor
technology. It only visualizes the resistance change over time on a big screen.

Decaying Signal 1 Decaying Signal 2

Logarithm of Decaying Signal Logarithm of Decaying Signal


with Linear Fit 1 with Linear Fit 2

Figure 8.5.: The developed user interface for decay time monitoring and the linear fit of the decaying
signal.

64
8.3. Testing

Decay time Capacitance


1.0

0.8

0.6

0.4

0.2

0.0

Resistance

Figure 8.6.: The developed user interface for capacitance and resistance monitoring.

8.3. Testing
8.3.1. Accuracy assessment of readout electronics
We wanted to assure that the arduino is capable of measuring such high resistances and small capacitances
accurately. Furthermore, we wanted to assess, which value of the measuring resistor (1, 2, 3 M ⌦) is most
suitable for the application. In order to measure this, we replaced the sensor by a varying number of 1M ⌦
resistors (1-10 M ⌦) in series with di↵erent capacitances (10, 22, 47, 100 pF). The data was acquired by
the scripts “1 Sensor functionality assessment.ino” and “1 Monitoring of decay Data acquisition.py” and
can be found in the folder “Electronics and Software/Sensor Readout one resistor/Electronics Evaluation”
together with the data processing script Plotting.py. Every data point was acquired 100 times and 25 and
75 percentiles were calculated and drawn for each one. The measured capacitance values as a function of
sensor resistance for the three di↵erent measurement resistors are displayed on the left of Fig. 8.7. The
applied capacitance value is plotted as a dotted black line. Due to parasitic capacitances of the setup
itself and production tolerances of the passive components the measured capacitance value was slightly
higher than the applied one. However, the readout remained constant for di↵erent resistances in a range
of 1-10 M ⌦. The measured resistance values as a function of the applied sensor resistance are displayed
on the right of Fig. 8.7. In an ideal case, they would all be straight lines (dotted black line). However,
especially for low sensor capacitances in combination with low sensor resistances and a high measurement
resistance, the curves displayed an o↵set. We assessed the approximate looprate (Fig. 8.8), which was
dependent on the sensor quantities and varied between 50 and 150 Hz.

65
8. Readout

(a) Measurement Resistance 1000 k⌦ (b) Measurement Resistance 1000 k⌦

(c) Measurement Resistance 2000 k⌦ (d) Measurement Resistance 2000 k⌦

(e) Measurement Resistance 3000 k⌦ (f) Measurement Resistance 3000 k⌦

Figure 8.7.: Accuracy assessment of the readout electronics. Left: Measured capacitance values as a
function of resistance. Right: Measured resistance values as a function of capacitance.

66
8.4. Discussion

(a) Measurement Resistance 1000 k⌦ (b) Measurement Resistance 2000 k⌦

(c) Measurement Resistance 3000 k⌦

Figure 8.8.: Loop rate of the readout electronics as a function of the sensor capacitance and resistance
for three di↵erent measurement resistors.

8.4. Discussion
In this chapter we developed a readout circuitry that is capable of measuring capacitances and resistances
accurately in the range of the sensor quantities. However, only a measurement resistance of 1 M ⌦ yields
a sufficient accuracy for the capacitance values later measured for the sensors (20-80 pF). For larger
measurement resistors the finite input inpedance of the ADC of the microcontroller seems to play a role.
The performance could be improved by using a recursive estimation approach (Kalman Filter) or a faster,
but bulkier, microcontrollers. If one wants also use the decay time or the resistance as the sensor output,
a small reservoir with ink, where the conductivity is constantly measured, would have to be attached
to the sensor. A di↵erential measurement to this reference ink reservoir would allow to compensate for
the temperature and humidity e↵ects that the environment would have on the resistive properties of the
ink. Finally, the readout electronics could be implemented on a flexible capton PCB with total area
requirement of 1- 2 cm2 .

67
9. Sensor Characterization
9.1. Introduction
In this chapter we characterize the sensors and compare them to the model that we developed in chapters
3 and 8. We first assess which of the connected sensors actually work and then present a failure mode
e↵ect analysis, where we discuss which fabrication steps were the most critical. Next, we compare the
static performance of the sensors to the model and subsequently move on to a dynamic characterization
on a tensile testing machine. Lastly, we present the mechanical characterization of the sensors as well as
of the electrical connection.

9.2. Functionality assessment of the sensors


9.2.1. Methodology
We tested their functionality by measuring their time response at two strains, namely 0 and 50 %.
We used the readout setup of the last chapter for data acquisition (scripts: “1 Sensor functionality
assessment.ino” and “1 Monitoring of decay Data acquisition.py”). The data can be found in folder
“Sensor characterization/SensorsFuncitonalityAssessment“.

9.2.2. Results
The sensors displayed the following four behaviors:
• Open circuit (OC)
• Short circuit (SC)
• Working sensor with one decay time
• Working sensors with two decay times at high strains
For each behavior the signal of the sensor at two strains and its linearization (Logarithm) was plotted .
The response of the open circuit is displayed in Fig. 9.1 and was manifested by a very short decay time
that was independent of strain. This corresponds to the charging parasitic capacitance of the readout
circuit. An open circuit occured, when the silver connections were insulated and did not make contact
with the conducting medium. The behavior of a short circuit is depicted in Fig. 9.2 and its main feature
was a persisting voltage signal at steady state. A short ciruit could occur, if there was a contact between
the two conducting layers, either via a hole at the end or in the middle of the elastomer, or if the wires
inside the elastomer touched each other. A working sensor that displayed one decay time for strains
up to 50 % is shown in Fig. 9.3. This sensor behaves as it was expected from the model. Sensor 32
displayed a behavior that consisted of two decay times at large strains (Fig. 9.4). This was manifested
in two straight lines separated with a kink in the linearized decay signal. We discovered that this was
actually the common mode of operation. Also for sensor 9, which exhibited only one decay time at 50 %,
a second decay time arose at strains of 200% and above (For three sensors, the linearized signal at 250
% displaying the kink between the two decay times can be found in appendix E.1.1 on page 111). The
problem could be circumvented by discarding the first half of the signal, as described in section 8.2.2.1 in
the last chapter. Figure 9.7 shows how the algorithm cropped the signal for sensor 25. From a practical
point of view the problem was solved, but we still investigated the cause of this behavior as described in
the next subsection. For completion, the response of the other sensors is shown in appendix E.1. It is to
say that sensor 3 worked in the functionality assessment, whereas sensor 6 did only partially. However,
when we did the static testing, sensor 3 (Fig. E.1 displayed an open loop response, whereas sensor 6 (Fig.
E.2) could be characterized. In summary, sensors 6, 8, 9, 14, 25 and 32 behaved as expected.

69
9. Sensor Characterization

(a) Raw Signal for 0 and 50 % strain. (b) Linearization of raw signal and linear fit

Figure 9.1.: Sensor 2 - Displaying an open circuit response

(a) Raw Signal for 0 and 50 % strain. (b) Linearization of raw signal and linear fit

Figure 9.2.: Sensor 12 - Displaying a short circuit response

(a) Raw Signal for 0 and 50 % strain. (b) Linearization of raw signal and linear fit

Figure 9.3.: Sensor 9 - Displaying the expected response with one decay time

70
9.2. Functionality assessment of the sensors

(a) Raw Signal for 0 and 50 % strain. (b) Linearization of raw signal and linear fit

Figure 9.4.: Sensor 32 - Displaying two decay times at higher strains

(a) Raw Signal (b) Linearization of raw signal

Figure 9.5.: Sensor 9 - Elongated up to a strain of 250 %, also starting to display a second decay time

9.2.2.1. Explanation for the two observed decay times in the decaying voltage signal
The first decay time did not change significantly with elongation (Fig. 9.6). We hypothize that a second
capacitor was charged. This capacitor was either smaller than the sensor capacitance or was of similar
size and had a lower resistance attached to it. Physical explanations for this could be the parasitic
capacitance of the readout circuitry or a capacitive coupling between the two silver wires via enclosed
ionic conductive spheres in the injected Dragonskin 30 at the end of the sensor. Both of these can be
described by the same equivalent readout circuitry, shown in Fig. 9.8a. The step response of this circuit
was simulated in Circuitlab 1 , the behavior of the modeled system paralleled experimental observation.
The simulated step response and its linearization are shown in Fig. 9.8b and 9.8c, respectively.

1 www.circuitlab.com, 30.07.2014

71
9. Sensor Characterization

(a) Linearization Sensor 8 (b) Linearization Sensor 25

Figure 9.6.: Sensor 8 and 25 highlighting that the first decay time remains constant as a function of strain
(slope before the kink remains the same)

Figure 9.7.: Sensor 25 with only half of the linearized signal.

72
9.3. Failure mode e↵ect analysis

Cs
Ro 10 pF
Ri
1M 1M

Rm
1M
C p,surr C p,i
Rp,i
20 pF 1.5 pF
2M

(a) Equivalent circuit of the sensor model together with a parasitic capaci-
tance in parallel to the sensors

(b) Simulated step response (c) Logarithm of simulated step response

Figure 9.8.: Simulation of equivalent circuit model for the explanation of two observed decay times.

9.2.3. Discussion
In this section we have identified six working sensors 6, 8, 9, 14, 25, 32 and given an explanation for
the two decay times observed. This problem can be circumvented by a proper algorithm. In the next
section, we discuss shortly what fabrication step the 34 sensors caused to fail in order to identify the
most promising approaches for further fabrication optimizations.

9.3. Failure mode e↵ect analysis


In this section we assess the impact of each fabrication step on sensor viability. Table 9.1 summarizes
the state of failure of the 34 sensors fabricated. The fabrication process was divided into three steps:
printing, wire connection and end cap fabrication (assembly). In printing, nine sensors failed, whereas
two problems occured. One was necking of the filament, which only occured in one case, and bubbles
also caused defective sensors (Fig. 9.9a). The bubbles were due to insufficient degassing of the inks.
Necking may have also been the result of air inclusion. In the wire connection, the main problem was the
injection of the Dragonskin, which was hard to control. Furthermore, it was difficult to displace all the
conductive liquid by elastomer. This resulted either in open or short circuits (Fig. 9.9b). In total, eight
sensors failed in this step. In the end cap fabriction, another seven sensors failed. The main failure was
either a pullout of the wire or a hole - both of which occured when doing the loop of the silver wire with
the needle.
To conclude, the printing and the assembly problems can both be solved easily through better degassing
and more training, respectively. The electrical connection will remain a problem though and is therefore

73
9. Sensor Characterization

considered the most critical step, where in subsequent work a better solution must be found.

Sensor Printing Printing Wire Con- Assembly Other Working


number Macroscopic Microscopic nection Process Sensors
1 Sensor
Lost
2 OC
3 OC
4 B
5 P
6 W
7 P
8 W
9 W
10
11 OC
12 SC
13 OC
14 W
15 H
16 MB
17 MB
18 MB
19 MB
20 MB
21 MB
22 SC
23 OC
24 MB
25 W
26 H
27 SC
28 N
29 H
30 SC
31 Not
finished
32 W
33 Not
finished
34 P
Total 2 7 9 6 3 6
B = Bubbles MB = OC = Open P = Wire W=
Microscopic Circuit Pullout Working
Bubble Sensor
N = Necking SC = Short H=
Circuit Hole/Air
inflow

Table 9.1.: Failure mode e↵ect analysis

74
9.4. Static testing

(a) Microscopic bubble (b) Short circuit, caused by the two silver wires touching

Figure 9.9.: Typical sensor failures

9.4. Static testing


9.4.1. Introduction
In this section we assess how well the sensor output and the model described in chapters 3 and 8 agree.

9.4.2. Methodology
For the static characterization we designed a testing board that allows the sensor to be strained to defined
values 2 . The tool consists of three parts that can be plugged into each other to be long enough for the
longest sensors. The data was acquired by an arduino over serial communication. The testing setup is
shown in Fig. 9.10. The scripts used were “2 Static readout half signal for linear fit.ino” and “4 Static
Data acquisition Sensors.py”. The six functioning sensors 6, 8, 9, 14, 25, 32 were evaluated. Except sensor
25, all sensors were strained to 250 % and data was acquired in reasonable intervals. For every data point
100 sensor property samples were acquired and averaged. Every sensor was elongated and retracted three
times. The maximum and minimum values for every data point are represented by the colored area
in the plots. All data and processing scripts can be found in the folder “Sensor characterization/Static
Sensor characterization/Data and plots”. Resistance and capacitance were plotted as a function of strain.
The model derived in chapter 8 was fit to the experimental data. The strain range over which the fit
was performed di↵ered from sensor to sensor and can be found in Tab. E.1. From the fit, the sensor
resistance Rs (equation 8.23), sensor capacitance Cs and parasitic capacitance Cp (equation 8.22) could
be determined. These quantities were then inserted into equation 8.26 to validate the model. This
expression was plotted together with the decay time data in a seperate plot for every sensor. We also
compared the experimentally determined quantities to the ones we obtained from the top down images
in chapter 6 and the values we measured for the conductivity (chapter 4). Gauge factor, defined as
C
C0
GF = L
(9.1)
L0

was calculated according to:


Cs
GF = (9.2)
Cp + Cs
where C0 = Cs + Cp and at 100 % strain C and L, that amounted C = Cs and L = L0 where
used to derive the last expression.

2 Thesolidworks files can be found in “Sensor characterization/Static Sensor characterization/Sensor Characterization


Tool”

75
9. Sensor Characterization

3
The maximum theoretical gauge factor for this sensor type is 1.

Strain testing board

Sensor

Data transfer to computer

Readout Electronics

(a) A sensor on the static characterization board to- (b) Sensors of di↵erent length on the calibration board.
gether with an arduino pro mini. Calculated The calibration board consisted of three such modules
data is transmitted via a serial port to the data that could be plugged into each other.
acquisition computer.

Figure 9.10.: Experimental setup for the static characterization of the sensors

9.4.3. Results
The results for the six sensors are displayed in Fig. 9.11 - 9.16. The experimental data showed good
agreement with the model over a large range of strain (0 - 250%). The best agreement was found for
sensors 14 and 32. For some sensors (9,14) there was a larger o↵set at low strains. A possible reason for this
could be that the inner tube was curvy because of the fabrication process and actually was not stretched
for low strains. In applications, this could be circumvented by prestraining the sensor. Data considerably
deviated from the model when the sensor resistance exceeded 14-15 M ⌦. The finite input impedance
of the readout electronics may have also lead to inaccuracies since it was not tested to accommodate
that large resistances. The determined sensor resistances and capacitances and the parasitic capacitances
for zero strain can be found in Tab. 9.2. They ranged from 3.78-20 pF for the sensor capacitance,
0.8-4.22 M ⌦ for the sensor resistance and 10.81 - 19.01 pF for the parasitic capacitance. This resulted
in gauge factors of 0.21 - 0.51 for the capacitance. The expected values from the calculations with the
data from the top down images were obtained by equations 3.5, 3.6 and is shown in Tab. 9.3. For the
conductivity, we used a value of 3 mS/cm (conductivity measurements) and for the dielectric constant a
value of ✏ = 2.5 [48]. Because we did not have cross section images of the sensor, we could not determine
the extent of the D-shape of the channel, hence we could not directly calculate the outer resistance.
However, we can make an estimation based on the ratio of the flowrates of the core and shell 2 (Tab.
6.4), which yield an area ratio of 5 (Sensors 1-8) and 3 (Sensors 9-34). Therefore an estimate of the total
sensor resistance could be determined and is given in Tab. 9.3. As we can see the order of magnitude
between the prediction and the actual value was correct. Except for sensor 25, the predicted values and
the measured ones showed the same trend. This was especially reflected in the high resistance values of
sensor 6 and 8, which were due to lower flowrates of the innermost channel during fabrication (Tab. 6.4).
Even more, if a 3 pF o↵set is added to the calculated capacitance values, accounting for the simplification
of the asymmetric geometry and uncertainties in the actual permittivity of the Dragonskin elastomer, we
obtained a stricking coincidence between experiment and prediction as show in Tab. 9.4. The predicted
resistance values are in four cases approximately 1.5 times higher than the measured ones, which, looking
at all the uncertainties in the calculation, was a very good result. The determined sensor quantities were
then inserted into equation 8.26 for the decay time and this quantity was plotted together with the data
to validate the model. As it can be seen in Fig. 9.11 - 9.16 the coincidence with the model was extremely
good. The decay time range was between 50-900 µs.

dC
3C 2·⇡·✏✏0 C 2·⇡·✏✏0 dC 2·⇡·✏✏0 dC C C
= ln(r2 /r1 )
L ! L
= ln(r2 /r1 )
and dL
= ln(r2 /r1 )
, therefore dL
= L
and hence GFid = dL =1
L

76
9.4. Static testing

(a) Resistance and capacitance as a function of strain to- (b) Decay time data together with the model prediction
gether with the values obtained through model fitting. with the values determined in the plot on the left.

Figure 9.11.: Sensor 6: Left: Determination of sensor properties, Right: Model validation

(a) Resistance and capacitance as a function of strain to- (b) Decay time data together with the model prediction
gether with the values obtained through model fitting. with the values determined in the plot on the left.

Figure 9.12.: Sensor 8: Left: Determination of sensor properties, Right: Model validation

77
9. Sensor Characterization

(a) Resistance and capacitance as a function of strain to- (b) Decay time data together with the model prediction
gether with the values obtained through model fitting. with the values determined in the plot on the left.

Figure 9.13.: Sensor 9: Left: Determination of sensor properties, Right: Model validation

(a) Resistance and capacitance as a function of strain to- (b) Decay time data together with the model prediction
gether with the values obtained through model fitting. with the values determined in the plot on the left.

Figure 9.14.: Sensor 14: Left: Determination of sensor properties, Right: Model validation

(a) Resistance and capacitance as a function of strain to- (b) Decay time data together with the model prediction
gether with the values obtained through model fitting. with the values determined in the plot on the left.

Figure 9.15.: Sensor 25: Left: Determination of sensor properties, Right: Model validation

78
9.4. Static testing

(a) Resistance and capacitance as a function of strain to- (b) Decay time data together with the model prediction
gether with the values obtained through model fitting. with the values determined in the plot on the left.

Figure 9.16.: Sensor 32: Left: Determination of sensor properties, Right: Model validation

Sensor Cs [pF] Cp [pF] Rs [M⌦] Gauge Factor


number
6 3.78 10.81 2.7 0.26
8 9.63 12.97 4.22 0.43
9 3.41 12.73 0.92 0.21
14 9.75 17.54 0.80 0.36
25 7.44 15.85 2.34 0.32
32 20.01 19.01 1.00 0.51

Table 9.2.: Overview of the experimentally determined sensor quantities.

Sensor Capacitance Resistance Resistance Total


number [pF] inner tube inner tube Resistance
[M⌦] [M⌦] [M⌦]
(estimated)
6 7.36 3.80 0.76 4.56
8 12.16 2.65 0.53 3.18
9 7.06 1.17 0.39 1.56
14 12.41 1.13 0.38 1.51
25 15.24 0.67 0.22 0.90
32 23.10 1.15 0.38 1.53

Table 9.3.: Overview of sensor quantities determined from the top down images. All the values are
an approximation since as we have seen in the chapter Readout, the values from top down
images were slightly too large. The capacitance values in particular, since in reality the
cylindrical capacitor is o↵-centered and we assumed a centered capacitor for the calculation.
The quantities used for the calculation can be found in Tab. C.3.

79
9. Sensor Characterization

Sensor Cs [pF] Cs top-down - 3pF [pF]


number
6 3.78 4.36
8 9.63 9.17
9 3.41 4.06
14 9.75 9.41
25 7.44 12.25
32 20.01 20.1

Table 9.4.: Measured capacitance and calculated capacitance from the top down images with the results
shifted by an amount of 3 pF.

9.4.4. Discussion
The static characterization has revealed that the sensors show an excellent agreement with the model.
The sensor properties were in the predicted range from the top-down images. However, as we have seen,
the model displayed large o↵sets for sensor resistances above 14-15 M ⌦. This suggests that further sensor
designs should be chosen in a way that the maximum sensor resistance remains below 10 M ⌦. We will
further discuss sensor optimization at the end of this chapter. The gauge factor can be easily improved,
by building an appropriate PCB with much less parasitic capacitances than the breadboard. In the next
section, we will assess the dynamic properties of the sensors.

9.5. Dynamic testing


9.5.1. Introduction
After the promising static testing we wanted to assess the sensors dynamic behavior. As determined
in chapter 8.3, depending on the sensor properties, the electronics has a bandwidth of 50-150 Hz. The
measured sensor properties will cause decay times and hence bandwidths of similar size. This leads to
a lower bound for resolving sinusoids of 25 Hz. In the next section we therefore want to assess whether
signals of a frequency lower than 25 Hz can be accurately resolved.

9.5.2. Methodology
We applied sawtooth strain signals to a sensor on an Instron 5544A Materials testing machine. All
data was acquired with sensor number 14. The sensor had an original length of 60 mm. In a typical
experiment, the sensor was prestrained by 20 mm at a velocity of 10 mm/s and then cycled 20 times at
a speed of 20 mm/s from 80 mm extension to 100 mm extension and then again retracted from 80 mm
to 60 mm at a speed of 10 mm/s. We explored di↵erent amplitudes (20 mm, 40 mm, 80 mm) and also
di↵erent speeds (5, 10, 20 mm/s). However, we only evaluated the experiments listed in table 9.5, which
correspond to actuation frequencies of 12 , 14 , 16 and 18 Hz. 4 The speed limit of the Instron 5544A of 25
mm/s did not allow us to explore much faster frequencies.
The experimental setup is shown in Fig. 9.17. The sensor readout was done by the Arduino and the
data transmitted serially to a computer via the scripts: “5 Dynamic Data acquisition Sensors.py” and
“4 Dynamic Readout half signal for linear fit dynamic measurements with synchronization channel.ino”.
The elongation reading was acquired by the Instron machine itself. Therefore, a synchronization of these
two signals was required. This was achieved by connecting an analog input of the Instron and a digital
input of the Arduino to the same row on the evaluation board and connecting this row shortly to 5V
right after the beginning of the experiment and then to ground again. A pulldown resistor of 10 k⌦ was
required that pulled the potential of the row to ground quickly as soon as the voltage supply was released.
This peak allowed the two signals to be synchronized when postprocessing the data. The shift between
the two signals was calculated by the maximum of their autocorrelation function and was done together
with the plotting in script “3 Paperplots dynamic measurements strain.py”, which can be found with

4 The other experiments are listed in table E.2 in appendix E.3.

80
9.5. Dynamic testing

all the data in the folder “Sensor characterization/DynamicSensorAnalysis Instron testing”. Before any
processing could be done, the two signals had to be transformed to the same sampling frequency. This
was done by a linear interpolation of the signals and a subsequent sampling at the desired frequency.
For half of the plots the signal was filtered by a second order butterworth filter (implemented in the
scipy library 5 ) with a cuto↵ frequency of 2 Hz. In order to perform the sensor calibration, all signals
were scaled and shifted by a linear transformation, which minimized the mean squared deviation from
the applied strain to sensor signal. This procedure was only fully adequate for the linear sensor output
capacitance and not for the nonlinear resistance and decay time signal. The plots still give an idea of
the noise levels associated with each sensor property. For comparison between di↵erent sensor properties
and the e↵ect of filtering, we calculated the mean absolute error defined as
n
1X
M AE = |ys (") "| (9.3)
n i=1

whereas ys (") is the sensor output and " the applied strain.

Figure 9.17.: Dynamic testing setup

Preelong- Prestrain [%] PeaktoPeak Maximum Velocity Frequency


ation [mm] [mm] strain [%] [mm/s] [Hz]
20 33.3% 20 66.6 % 20 0.5
20 33.3% 40 100 % 20 0.25
20 33.3% 80 166.6 % 20 0.125
20 33.3% 80 166.6 % 10 0.0625

Table 9.5.: Actuation amplitude and frequency for the evaluated dynamic tests. All tests were performed
by sensor 14.

9.5.3. Results
Figure 9.18 shows the sensor output for the capacitance reading during the dynamic test. The unfiltered
signal was very noisy, but after filtering, the capacitance data tracked the applied strain signal very well.
An o↵set was observed at low strains. This error is explainable if one considers Fig. 9.14, which displays
a deviation from the sensor model at low strains, probably because the inner tube of the sensor at the
beginning was curvy. This could be handled by applying a larger prestrain. The resistance reading was

5 http://www.scipy.org/

81
9. Sensor Characterization

already less noisy (Fig. 9.19), but the nonlinear sensor response remained. Here a calibration curve
would have to be obtained first and then the exact strain values would have to be calculated by a linear
interpolation from these data points. The best SNR-ratio displayed the decay time signal (Fig. 9.20), but
a calibration curve would have to be stored and linear interpolation to be performed for a linear output of
the sensor. Furthermore, resistance and decay time will be subject to temperature and humidity changes,
since they alter the conductivity properties of the ionic conductor. A way to overcome this was mentioned
in section 8.4.

(a) Unfiltered Signal (b) Filtered Signal

Figure 9.18.: Sensor output and applied strain for di↵erent frequencies, when the sensor capacitance was
used as the readout signal. Left: Unfiltered Signal, Right: Filtered Signal

(a) Unfiltered Signal (b) Filtered Signal

Figure 9.19.: Sensor output and applied strain for di↵erent frequencies, when the sensor resistance was
used as the readout signal. Left: Unfiltered Signal, Right: Filtered Signal

82
9.5. Dynamic testing

(a) Unfiltered Signal (b) Filtered Signal

Figure 9.20.: Sensor output and applied strain for di↵erent frequencies, when the decay time was used as
the readout signal. Left: Unfiltered Signal, Right: Filtered Signal

The calculated mean absolute errors are displayed in table 9.6, 9.7 and 9.8. The errors increased with the
applied strain amplitude. For the unfiltered signal, errors were largest for the capacitance reading and
smallest for the decay time reading, despite the decay time signal is actually not linear. After filtering,
the data suggested that the decay time signals was the most accurate, since for low strains its error was
smaller than the one of the capacitance signal. For larger strain amplitudes the non-linearity played a
more important role, which can be seen easily since the filtering did not significantly improve the mean
absolute error for resistance and decay time at large strains.

MAE MAE Strain min Strain max Frequency [Hz]


Unfiltered Filtered [% [%] [%]
[% strain] strain]
5.6 % 0.9 % 33.3 % 66.6 % 0.5
8.2 % 2.4 % 33.3 % 100 % 0.25
11.5 % 4.3 % 33.3 % 166.6 % 0.125
11.2 % 4.1 % 33.3 % 166.6 % 0.0625

Table 9.6.: Mean absolute error for the capacitance readings at di↵erent amplitudes and frequencies.

MAE MAE Strain min Strain max Frequency [Hz]


Unfiltered Filtered [% [%] [%]
[% strain] strain]
6.0 % 2% 33.3 % 66.6 % 0.5
5.7 % 3.5 % 33.3 % 100 % 0.25
6.4 % 5.6 % 33.3 % 166.6 % 0.125
6.1 % 5.3 % 33.3 % 166.6 % 0.0625

Table 9.7.: Mean absolute error for the resistance readings at di↵erent amplitudes and frequencies.

83
9. Sensor Characterization

MAE MAE Strain min Strain max Frequency [Hz]


Unfiltered Filtered [% [%] [%]
[% strain] strain]
2.4 % 0.7 % 33.3 % 66.6 % 0.5
2.8 % 2.1 % 33.3 % 100 % 0.25
5.8 % 5.5 % 33.3 % 166.6 % 0.125
5.6 % 5.3 % 33.3 % 166.6 % 0.0625

Table 9.8.: Mean absolute error for the decay time readings at di↵erent amplitudes and frequencies.

9.5.4. Discussion
We have demonstrated that the sensor is capable of tracking signals at low frequencies accurately. The
lowest SNR were found for the decay time reading, but this is a non-linear sensor. Future work will
therefore try to increase the SNR of the capacitance reading, by optimizing the sensor fabrication to
produce a smaller dielectric layer, hence creating a larger capacitance. We will discuss an approach for
optimization at the end of this chapter. But first, we will assess the mechanical properties of the sensors
in the next section.

9.6. Mechanical testing


The purpose of the mechanical testing was to determine the elongation at break and the pullout force of
the connection cables.

9.6.1. Elongation at break


The elongation at break was determined by a Instron 5544A tensile testing machine. Elongation rate was
set to 5mm/s. Three sensors of length 30, 32, 33 mm were tested until failure. Two failed at the interface
of the sensor tube with the end caps and one failed in the middle of the tube. The strains at failure can
be found and the load at failure can be found in Tab 9.9. On average the sensor tore at 770 % ± 37 %
at a load of 3.24 N ± 0.18 N. Load/strain curves can be found in appendix E.3.1 Fig. E.13.

Sensor Load at Break [N] Strain at break [%]


a 3.42 822
b 3.00 742
c 3.30 744

Table 9.9.: Values for the load and strain at break

9.6.2. Connection cable strength


The connection cable strength was assessed on a Instron 5544A tensile testing machine. Tests were
performed at an elongation rate of 0.5 mm/s. The two connectors were tested individually. In total, the
strengths of four connection cables were assessed. Values determined for the pullout forces can be found
in Tab 9.10. On average the connections failed at a force of 1.47 N ± 0.1534 N. The load/elongation
curves can be found in appendix E.3.1 Fig. E.12.

Specimen Pullout force [N]


1 1.71
2 1.28
3 1.45
4 1.47

Table 9.10.: Values for the pullout forces of the connections.

84
9.7. Discussion sensor characterization

9.7. Discussion sensor characterization


In the last few sections we have performed a lot of characterizations on the sensors. In the static
testing section, experimental results were in excellent agreement with analytical predictions. The sensor
resistance should be kept below 10 M ⌦ in order for the sensors not to display a large deviation from
the model. In the dynamic testing section, we observed that the sensors were able to track a signal
accurately and that they did not display a hysteresis. However, the resistance of the sensors was still
very high (1-4 M⌦) and the capacitance (below 20 pF ) very low. This resulted in a very low SNR ratio
for the capacitance reading, and a high degree of filtering that was required in order to track the signal
in the dynamic measurements. In chapter 2.2.1 we identified the constraint on the sampling frequency
of a sensor of length 30 mm for measuring knee angles at an accuracy of ± 5 deg to be 110 Hz. With
our readout velocity of 50-150 Hz we obtained an accuracy of ± 3.5 - 11 deg, which meets the design
specifications. However, the maximum sensor decay time measured was approximately 900 ms (Sensor
32 Figure 9.16) and therefore the constraint on the bandwidth imposed by the physical properties of
this sensor will be approximately 1 kHz therefore allowing sinusoid of up to 500 Hz to be measured, if
one assumes that the readout electronics is infinitively fast. We have not explicitly measured the cross
sensitivity to pressure but this is certainly a point for further investigation. A hydrogel soaked with
glycerol may decrease sensor cross-sensitivity as mentioned at the end of chapter 4. The sensors are
very strechable (770 %) and the loop in the connection cable increased the pullout force to 1.47 N per
connector. A force at break of 3.24 N indicates that the sensor itself will fail prior to wire pullout. We
now present an outlook how sensor performance could be optimized in the future.

9.7.1. Proposed approach for sensor optimization


The figure of merit for a good sensor performance are the three sensor properties ⌧ , Rs and Cs . In terms
to obtain a high bandwidth, the decay time should be kept as low as possible. However, if it is too low
the signal decays too quickly for the ADC in order to resolve it. The sensor capacitance should be as
high as possible (Figure 3.4) and the sensor resistance below 10 M ⌦ for proper sensor operation. With
optimizations of the nozzle design, the dielectric thickness of the capacitor can be decreased to 30-40
µm. Therefore, the whole sensor design problem can be formulated as an optimization problem with
constraints on:

• The maximal sensor radius rsens = r40


• The minimal outer wall thickness wmin = r40 r30

• The sensor length


• The maximal strain

• The dielectric diameter, limited by the fabrication process to 30-40 µm, r20 r10 = 30µm

• The maximal sensor resistance of 10 M ⌦.

• The desired decay time range, for the current sampling frequency of 75 kHz the decay time must
not be below 50 µs.
Many of these constraints come from the actual design problem, which will have to be solved for every new
design. The output of this analysis will be the optimal ratio of the four radii under the given constraints.
We give an example how the maximum strain determines the maximum initial sensor resistance (Equation
9.4) and via that gives a constraint on the three internal radii (Equation 9.5).
Depending on the magnitude of the applied strain, the initial sensor resistance can be calculated via a
rearrangement of equation 8.23 to
10M ⌦
Rs0 ("max ) = (9.4)
("2max + 2"max + 1)
For example, for a maximum strain of 500% the initial sensor resistance should be below 270 k⌦. The
design parameters of the sensor are then given by

85
9. Sensor Characterization

L0
✓ ◆
2 1 1
Rs0 = ⇢ + 2 (9.5)
⇡ (r30 2 r20 2 ) r10
whereas ⇢ was determined to be around 2-3 mS/cm (section 4.4). For a known length, we have obtained
a constraint on the three radii. As rule of thumb, the value of the two resistances should be chosen to
be similar, in order to have a resistance that is as low as possible and to increase the capacitance by a
large r1 (6.4). In the future, a software should be developed, that does this optimizations for the di↵erent
input parameters and calculates the required sensor design parameters is a key requirement in order to
quickly adapt sensor design to varying material properties or new fabrication technologies. After having
outlined an optimization, we present an application for joint angle monitoring in the next chapter.

86
10. Applications
10.1. Introduction
The soft ionic strain sensors were originaly designed to be directly integrated into a piece of fabric. For
this reason, we dedicate the last chapter to a demonstration of ait analysis by a sensor attached to the
knee region of a flexible pants.

10.2. Methodology
Sensor number 6 was sewn onto a elastic pant via the micromolded attachement holes of the sensor end
cap. The readout electronics was put into a pocket of the jacket. We walked at di↵erent speeds on a
treadmill (1, 2, 3, 4 miles per hour). Fig. 10.1 shows the testing setup. The attachement is shown
enlarged in Fig. 10.1b. The scripts used for data acquisition and visualization were discussed in chapter
8. The data was low-pass filtered after aquisition. All raw data and processing scripts can be found in
folder Applications/Treadmill walking.

(a) The setup for acquisition of walking data (b) The sensor attachement to the fab-
ric.

Figure 10.1.: The setup for the walking monitoring with a soft strain sensor attached to the knee.

10.3. Results
Fig. 10.2 shows the normalized decay time data of the sensor output, while walking at di↵erent speeds.
One can see that for faster walking speeds a second peak arises, which is in coincidence with biomedical
data of addititional knee swinging of humans walking at faster speeds [58].

87
10. Applications

d)
1 mph

2 mph

0 3 mph

4 mph

Time [s]

Figure 10.2.: Normalized decay time data for walking at di↵erent speeds.

A movie with a live demo of how the sensor output changes under knee flexion is in the folder “Videos/Knee
angle measurements cut.mp4”.

10.4. Discussion
The sensors can resolve human walking patterns even with the current readout setup and could in principle
be used for feedback control of a soft exoskeleton. One can easily go from externally attached sensors to
fully integrated sensors (Fig. 10.3). We will now conclude and summarize the results and contributions
of this work.

Figure 10.3.: A fully integrated sensor into a fabric that could sense the movement of the wrist. The
concept can be easily extended to individual finger sensing.

88
11. Conclusion and Outlook
11.1. Summary
The goal of this thesis was to develop a proof of concept soft strain sensor for joint angle monitoring in a
soft exosuit. Within the deadline of this thesis we have come up with a robust, textile integrated sensors
that meets the temporal resolution requirements that we have specified in chapter 2. The readout is
simple and the implementation on the Arduino renders it a low power device. The sensor is comfortable
to wear and conforms easily to any curved surface. In further iterations it could even be rendered machine
washable. The sensor is easy to fabricate and with appropriate automation can be scaled up for mass
fabrication by a 3D printing process. The risks associated with evaporation were solved by using a non-
volatile solvent and the low conductivity turned out not to be as important as originially thought. The
cross sensitivity to pressure, although not quantified, is still an issue and ways to overcome it have been
mentionned in chapter 4. Incompatibilities between the materials did not occur. The interface between
hard and soft materials was very robust and one of the mechanically tested sensors actually broke in the
middle and not at the interface.

11.2. Significance
A linear, capacitive, hysteresis free strain sensor operating at 250 % strain has not been demonstrated to
date. The design of the electrical connection with the loop in the cable is very robust and solved one of
the most challenging problems in soft electronics, namely the combination of soft and hard materials in
the same device. It is a versatile element that can be fabricated at any length and integrated in almost
any structure through a simply weaving or molding process. In future iterations, the sensor could be
integrated into the soft exoskeleton and provide real time joint angle data for feedback control of the
suit. The sensors may impact other fields as well. For instance, a suit could be developed that track
the whole body posture of the player and conveys its movements into the virtual world, allowing for a
more realistic game experience. Health monitoring systems could also be developed. For instance, shirts
with integrated sensors that monitor sitting posture, could remind office people to adapt their posture
in order to avoid back problems. In sports, such a sensor suit could be used to better understand how
experts perform their movements. The printing technology itself could pave the way for new composite
or functional materials.

11.3. Outlook
Future work is possible in sensor design, materials, fabrication, electronics, and software. More precise
nozzles could be fabricated with pulled glass tubes by a pop-up MEMS process as described in chapter
5. The ratio and absolute values of the four layers could be adjusted to maximize sensitivity. A smaller
printing setup with four stepper motors/syringe pairs embedded into the print head would greatly improve
the printing process. Furthermore, UV-curable elastomers could be investigated, which would in principle
allow to fabricate a continuous sensor filament and to collect it on a reel. With the 3D printing setup,
more complicated layered patterns like 2D sensor arrays could be investigated. The end caps of the
sensors could be made smaller to yield more threadlike sensors. One could explore methods to directly
bond the sensors onto fabric. Multiple sensors could be used in parallel as an array, which would allow for
the isolation of high pressure points. In terms of materials, one could examine less elastic and more robust
materials like polyurethanes, since most often only 20 % strain is required in applications. The cross
sensitivity to pressure could be reduced by exploring an elastic ionic conductor instead of a fluid. The
dielectric constant of the capacitor could be increased by incorporating nanopowders of materials with a
high dielectric constant. The sensors limitation at higher frequencies should be investigated as well as the

89
11. Conclusion and Outlook

long term robustness and cycles to failure. The entire readout electronics could be designed on a flexible
PCB and directly integrated into the fabric together with the sensor. Lastly, also other applications areas
are possible. For instance, implantability could be easily achieved by using a ionic conductor based on
simulated body fluids, which would pave the way to soft medical devices with proprioceptive feedback.
The multicore shell printing technology might find applications in tissue enginnering and 3D printing of
structural materials.

90
Bibliography
[1] Wei Zeng, Lin Shu, Qiao Li, Song Chen, Fei Wang, and Xiao-Ming Tao. Fiber-Based Wearable
Electronics: A Review of Materials, Fabrication, Devices, and Applications. Advanced materials
(Deerfield Beach, Fla.), June 2014.
[2] Leif P Jentoft, Student Member, Yaroslav Tenzer Member, Daniel Vogt Member, Jia Liu, Robert
J Wood Member, and Robert D Howe Fellow. Flexible , Stretchable Tactile Arrays From MEMS
Barometers.
[3] Panagiotis Polygerinos, Stacey Lyne, Luis Fernando Nicolini, Bobak Mosadegh, George M. White-
sides, and Conor J. Walsh. Towards a soft pneumatic glove for hand rehabilitation. 2013 IEEE/RSJ
International Conference on Intelligent Robots and Systems, pages 1512–1517, November 2013.
[4] By Rolf Pfeifer, Max Lungarella, and Fumiya Iida. The Challenges ahead for bio-inspired soft
robotics. Communications of the ACM, 55(11), 2012.

[5] R. Morales, F. J. Badesa, N. Garcia-Aracil, J. M. Sabater, and L. Zollo. Soft Robotic Manipulation
of Onions and Artichokes in the Food Industry. Advances in Mechanical Engineering, 2014:1–9,
2014.
[6] C Edwards and R Marks. Evaluation of biomechanical properties of human skin. Clinics in derma-
tology, 13(4):375–80.

[7] Yigit Menguc, Yong-Lae Park, Ernesto Martinez-Villalpando, Patrick Aubin, Miriam Zisook, Leia
Stirling, Robert J. Wood, and Conor J. Walsh. Soft wearable motion sensing suit for lower limb
biomechanics measurements. 2013 IEEE International Conference on Robotics and Automation,
pages 5309–5316, May 2013.

[8] Lijia Pan, Alex Chortos, Guihua Yu, Yaqun Wang, Scott Isaacson, Ranulfo Allen, Yi Shi, Reinhold
Dauskardt, and Zhenan Bao. An ultra-sensitive resistive pressure sensor based on hollow-sphere mi-
crostructure induced elasticity in conducting polymer film. Nature communications, 5:3002, January
2014.

[9] Daniel Vogt, Yong-Lae Park, and Robert J. Wood. A soft multi-axis force sensor. 2012 IEEE
Sensors, pages 1–4, October 2012.

[10] Yong-Lae Park, Bor-rong Chen, and Robert J. Wood. Soft artificial skin with multi-modal sensing
capability using embedded liquid conductors. 2011 IEEE SENSORS Proceedings, 2:81–84, October
2011.
[11] Peter Roberts, Dana D. Damian, Wanliang Shan, Tong Lu, and Carmel Majidi. Soft-matter capac-
itive sensor for measuring shear and pressure deformation. 2013 IEEE International Conference on
Robotics and Automation, pages 3529–3534, May 2013.

[12] Lucie Viry, Alessandro Levi, Massimo Totaro, Alessio Mondini, Virgilio Mattoli, Barbara Mazzolai,
and Lucia Beccai. Flexible Three-Axial Force Sensor for Soft and Highly Sensitive Artificial Touch.
Advanced materials (Deerfield Beach, Fla.), pages 1–6, February 2014.
[13] Waris Obitayo and Tao Liu. A Review: Carbon Nanotube-Based Piezoresistive Strain Sensors.
Journal of Sensors, 2012:1–15, 2012.
[14] Christian a Gutierrez and Ellis Meng. Low-cost carbon thick-film strain sensors for implantable
applications. Journal of Micromechanics and Microengineering, 20(9):095028, September 2010.

91
Bibliography

[15] Minwoo Park, Jungkyun Im, Minkwan Shin, Yuho Min, Jaeyoon Park, Heesook Cho, Soojin Park,
Mun-bo Shim, Sanghun Jeon, Dae-young Chung, Jihyun Bae, Jongjin Park, Unyong Jeong, and
Kinam Kim. Highly stretchable electric circuits from a composite material of silver nanoparticles
and elastomeric fibres. 7(December), 2012.
[16] Sude Ma, Yan Wang, Zhonghua Min, and Lisheng Zhong. Intrinsically conducting polymer-based
fabric strain sensors. Polymer International, (April):n/a–n/a, April 2013.
[17] Yi Wang, Rong Yang, Zhiwen Shi, Lianchang Zhang, Dongxia Shi, Enge Wang, and Guangyu Zhang.
Super-elastic graphene ripples for flexible strain sensors. ACS nano, 5(5):3645–50, May 2011.
[18] Darren J Lipomi, Michael Vosgueritchian, Benjamin C-K Tee, Sondra L Hellstrom, Jennifer a Lee,
Courtney H Fox, and Zhenan Bao. Skin-like pressure and strain sensors based on transparent elastic
films of carbon nanotubes. Nature nanotechnology, 6(12):788–92, December 2011.
[19] Christoph Keplinger, Jeong-Yun Sun, Choon Chiang Foo, Philipp Rothemund, George M White-
sides, and Zhigang Suo. Stretchable, transparent, ionic conductors. Science (New York, N.Y.),
341(6149):984–7, August 2013.
[20] Joseph T Muth, Daniel M Vogt, Ryan L Truby, Yigit Menguc, David B Kolesky, Robert J Wood, and
Jennifer a Lewis. Embedded 3D Printing of Strain Sensors within Highly Stretchable Elastomers.
Advanced materials (Deerfield Beach, Fla.), June 2014.
[21] a. Lion and C. Kardelky. The Payne e↵ect in finite viscoelasticity. International Journal of Plasticity,
20(7):1313–1345, July 2004.
[22] Shouli Wang, Peng Wang, and Tianhuai Ding. Resistive Viscoelasticity of Silicone Rubber / Carbon
Black Composite. 2011.
[23] Shu Zhu, Ju-Hee So, Robin Mays, Sharvil Desai, William R. Barnes, Behnam Pourdeyhimi, and
Michael D. Dickey. Ultrastretchable Fibers with Metallic Conductivity Using a Liquid Metal Alloy
Core. Advanced Functional Materials, 23(18):2308–2314, May 2013.
[24] Yong-Lae Park, Carmel Majidi, Rebecca Kramer, Phillipe Bérard, and Robert J Wood. Hyperelastic
pressure sensing with a liquid-embedded elastomer. Journal of Micromechanics and Microengineer-
ing, 20(12):125029, December 2010.
[25] Ming-Yuan Cheng, Chun-Liang Lin, Yu-Tse Lai, and Yao-Joe Yang. A polymer-based capacitive
sensing array for normal and shear force measurement. Sensors (Basel, Switzerland), 10(11):10211–
25, January 2010.
[26] C Majidi, R Kramer, and R J Wood. A non-di↵erential elastomer curvature sensor for softer-than-
skin electronics. Smart Materials and Structures, 20(10):105017, October 2011.
[27] Tong Lu, Lauren Finkenauer, James Wissman, and Carmel Majidi. Rapid Prototyping for Soft-
Matter Electronics. Advanced Functional Materials, pages n/a–n/a, February 2014.
[28] Prakash Manandhar, Paul D. Calvert, and John R. Buck. Elastomeric Ionic Hydrogel Sensor for
Large Strains. IEEE Sensors Journal, 12(6):2052–2061, June 2012.
[29] Yin-Nee Cheung, Yun Zhu, Ching-Hsiang Cheng, Chen Chao, and Wallace Woon-Fong Leung.
A novel fluidic strain sensor for large strain measurement. Sensors and Actuators A: Physical,
147(2):401–408, October 2008.
[30] Jean-Baptiste Chossat, Yong-Lae Park, Robert J. Wood, and Vincent Duchaine. A Soft Strain
Sensor Based on Ionic and Metal Liquids. IEEE Sensors Journal, 13(9):3405–3414, September 2013.
[31] Andrew H Gosline, Veaceslav Arabagi, Adil Kassam, and Pierre E Dupont. Achieving Biocompati-
bility in Soft Sensors for Surgical Robots. pages 5–6, 2013.
[32] Kyoung-Yong Chun, Youngseok Oh, Jonghyun Rho, Jong-Hyun Ahn, Young-Jin Kim, Hyouk Ryeol
Choi, and Seunghyun Baik. Highly conductive, printable and stretchable composite films of carbon
nanotubes and silver. Nature nanotechnology, 5(12):853–7, December 2010.

92
Bibliography

[33] Bong Sup Shim, Wei Chen, Chris Doty, Chuanlai Xu, and Nicholas a Kotov. Smart electronic yarns
and wearable fabrics for human biomonitoring made by carbon nanotube coating with polyelec-
trolytes. Nano letters, 8(12):4151–7, December 2008.

[34] J H Lee, D Yang, S Kim, and I Park. STRETCHABLE STRAIN SENSOR BASED ON METAL
NANOPARTICLE THIN FILM FOR HUMAN MOTION DETECTION & FLEXIBLE PRESSURE
SENSING DEVICES Korea Advanced Institute of Science and Technology , Republic of Korea
Fabrication of Stretchable Strain Sensors based on Silver. 1(June):2624–2627, 2013.

[35] David J Lorang, Douglas Tanaka, Christopher M Spadaccini, Klint a Rose, Nerine J Cherepy, and
Jennifer a Lewis. Photocurable liquid core-fugitive shell printing of optical waveguides. Advanced
materials (Deerfield Beach, Fla.), 23(43):5055–8, November 2011.

[36] Robert Brooke, Drew Evans, Maik Dienel, Pejman Hojati-Talemi, Peter Murphy, and Manrico Fab-
retto. Inkjet printing and vapor phase polymerization: patterned conductive PEDOT for electronic
applications. Journal of Materials Chemistry C, 1(20):3353, 2013.

[37] Yao Cheng, Fuyin Zheng, Jie Lu, Luoran Shang, Zhuoying Xie, Yuanjin Zhao, Yongping Chen, and
Zhongze Gu. Bioinspired Multicompartmental Microfibers from Microfluidics. Advanced materials
(Deerfield Beach, Fla.), pages 1–7, June 2014.
[38] Q. Hu. Rubber composite fibers containing silver nanoparticles prepared by electrospinning and
in-situ chemical crosslinking. Express Polymer Letters, 6(4):258–265, February 2012.
[39] Manuel J Dı́az-de León. Electrospinning Nanofibers of Polyaniline and Polyaniline / ( Polystyrene
and Polyethylene Oxide ) Blends Abstract. 1:1–5, 2001.
[40] Corinne Mattmann, Frank Clemens, and Gerhard Tröster. Sensor for Measuring Strain in Textile.
Sensors, 8(6):3719–3732, June 2008.

[41] R.Canegallo M.Sergio, N.Manaresi, M.Tartagni, R.Guerrieri. 5 . 4 : A Textile Based Capacitive


Pressure Sensor. pages 1625–1630, 2002.
[42] Alan T Asbeck, Robert J Dyer, Arnar F Larusson, and Conor J Walsh. Biologically-inspired soft
exosuit. IEEE ... International Conference on Rehabilitation Robotics : [proceedings], 2013:1–8, June
2013.
[43] Ye Ding, Ignacio Galiana, Alan Asbeck, Brendan Quinlivan, Stefano Marco, and Maria De Rossi.
Multi-joint Actuation Platform for Lower Extremity Soft Exosuits.
[44] J C Cool. Biomechanics of orthoses for the subluxed shoulder. Prosthetics and orthotics international,
13(2):90–6, August 1989.
[45] Frederick F Ling. Electromechanical Sensors and Actuators. Springer Berlin Heidelberg, 1999.
[46] G.S.N. RAJU. Electromagnetic Field Theory and Transmission Lines. Dorling Kindersley, 2006.
[47] H. Alexander. A constitutive relation for rubber-like materials. International Journal of Engineering
Science, 6(9):549 – 563, 1968.
[48] J.E. Mark. Polymer Data Handbook. Oxford University Press, Incorporated, 1999.
[49] Jennifer A. Lewis. Direct Ink Writing of 3D Functional Materials. Advanced Functional Materials,
16(17):2193–2204, November 2006.

[50] Michael Freemantle. An Introduction to Ionic Liquids. RSC Publishing, 2010.


[51] Ortrud Aschenbrenner, Somsak Supasitmongkol, Marie Taylor, and Peter Styring. Measurement
of vapour pressures of ionic liquids and other low vapour pressure solvents. Green Chemistry,
11(8):1217, 2009.
[52] Your universally applicable Polymer. Clariant, Edition 2007.

93
Bibliography

[53] Vincent A. Hackley. Guide to Rheological Nomenclature: Measurements in Ceramic Particulate


Systems, volume NIST Special Publication 946. NIST, 2001.

[54] Glycerine:an overview. The Soap and Detergent Association, 1990.


[55] Jeong-Yun Sun, Xuanhe Zhao, Widusha R K Illeperuma, Ovijit Chaudhuri, Kyu Hwan Oh, David J
Mooney, Joost J Vlassak, and Zhigang Suo. Highly stretchable and tough hydrogels. Nature,
489(7414):133–6, September 2012.
[56] J P Whitney, P S Sreetharan, K Y Ma, and R J Wood. Pop-up book mems. Journal of Microme-
chanics and Microengineering, 21(11):115021, 2011.

[57] Reinhard Lerch. Elektrische Messtechnik. Springer, 2012.


[58] Michael H Schwartz, Adam Rozumalski, and Joyce P Trost. The e↵ect of walking speed on the gait
of typically developing children. Journal of biomechanics, 41(8):1639–50, January 2008.

94
A. Sensing Material Development
A.1. Printing of Core/Shell eGaIn/Ecoflex wires

Figure A.1.: eGaIn core printed into a Ecoflex® 00-30 shell. The diameter of the extruded line is around
1mm thick. It can be seen that the conductive core formed spheres due to the high surface
tension of the conductive metal.

Figure A.2.: Conductivity transients due to the high viscosity of the ink for the di↵erent conductivity
measurements.

95
B. Nozzle Fabrication Data

Gauge Inner Diameter [mm] OuterDiameter [mm]


7 3.81 4.191
9 3.2512 3.7592
11 2.54 3.048
13 1.9558 2.413
15 1.524 1.8288
17.5 1.1684 1.4224
19 0.889 1.0668
20.5 0.6858 0.8636
23 0.4318 0.635
27 0.254 0.4064
32 0.1016 0.2286

Table B.1.: Tubing sizes overview.

96
Serial Number Core Shell 1 Shell 2 Shell 3 Core Length Shell 1 Shell 2 Shell 3 Generation
Number of Layers [Gauge] [Gauge] [Gauge] [Gauge] [mm] Length [mm] Length [mm] Length [mm]
1 2 23 19 0 0 28 16 0 0 2
2 2 27 20.5 0 0 28 16 0 0 3
3 2 17.5 13 0 0 28 16 0 0 3
4 2 27 13 0 0 28 16 0 0 3
5 3 23 19 15 0 40 28 16 0 3
6 2 23 19 0 0 28 16 0 0 3
7 4 23 19 15 11 52 40 28 16 3
8 4 27 20.5 17.5 13 52 40 28 16 3
9 2 17.5 13 0 0 28 16 0 0 3
10 3 27 20.5 17.5 0 40 28 16 0 3
11 4 32 23 19 15 16 16 16 16 4
12 4 27 20.5 17.5 13 16 16 16 16 4
13 2 20.5 17.5 0 0 16 16 0 0 4
14 2 23 19 0 0 16 16 0 0 4
15 2 23 19 0 0 16 16 0 0 4

Table B.2.: Tubing sizes of the core and shells of the farbricated nozzles
97
98
B. Nozzle Fabrication Data

Serial IM Layer1 IM Layer2 IM Layer3 IM Layer1 IM Layer2 IM Layer3


Number [Gauge] [Gauge] [Gauge] Length [mm] Length [mm] Length [mm]
1 20.5 0 0 18 0 0
2 23 0 0 18 0 0
3 15 0 0 18 0 0
4 0 0 0 0 0 0
5 20.5 17.5 0 30 18 0
6 20.5 0 0 18 0 0
7 20.5 17.5 13 42 30 18
8 23 19 15 42 30 18
9 15 0 0 18 0 0
10 23 19 0 30 18 0
11 27 20.5 17.5 42 30 18
12 23 19 15 42 30 18
13 19 0 0 18 0 0
14 20.5 0 0 18 0 0
15 20.5 0 0 18 0 0

Table B.3.: Tubing sizes of the intermediate layers of the fabricated nozzles
C. Printing Process Data

Speed Core [µm] Shell 1 [µm] Shell 2 [µm] Shell 3 [µm] Dielectric Dielectric
[mm/s] small [µm] large [µm]
2 589 1012 1700 2098 120 287
2 583 994 1702 2095 117 310
2 567 970 1703 2088 121 290
2 592 970 1708 2093 105 273
2 586 991 1693 2100 116 305
4 441 756 1287 1555 77 192
4 438 729 1245 1558 100 184
4 442 716 1296 1524 103 169
4 419 674 1273 1512 92 175
4 432 723 1286 1524 90 165
5 341 613 1168 1381 127 150
5 339 614 1164 1361 112 135
5 399 662 1165 1346 112 155
5 385 650 1164 1351 114 133
5 395 628 1123 1305 99 146

Table C.1.: Dragonskin 10 speed test raw data from top down microscopic images

99
100
C. Printing Process Data

Speed Core Shell 1 Height Width EncapsulationEncapsulationO↵set Dielectric Dielectric Dielectric Dielectric Dielectric
[mm/s] [µm] [µm] Shell 2 Shell 2 botton top [µm] dielec- average mini- maxi- small large
[µm] [µm] [µm] tric [µm] mum mum [µm] [µm]
[µm] [µm] [µm]
2 512 776 1073 1384 330 217 68 132 64 200 76 252
2 438 736 1025 1359 321 247 60 149 89 209 90 249
2 430 708 1032 1391 382 210 71 139 68 210 77 238
2 464 714 1003 1432 315 206 51 125 74 176 82 276
2 472 690 1027 1411 247 202 58 109 51 167 64 258
4 336 522 705 948 213 185 44 93 49 137 50 142
4 306 518 716 946 226 205 48 106 58 154 61 154
4 328 528 716 964 200 178 59 100 41 159 48 160
4 352 552 743 959 212 187 49 100 51 149 52 171
4 316 544 727 930 206 176 43 114 71 157 71 178
5 288 436 595 857 152 162 32 74 42 106 40 103
5 294 450 638 856 176 151 39 78 39 117 50 126
5 352 520 657 888 162 175 42 84 42 126 52 138
5 356 532 596 810 154 167 40 88 48 128 48 113
5 294 498 605 833 185 148 34 102 68 136 42 103

Table C.2.: Dragonskin speed test raw data from crosssection images.
(a) Cross section highlighting how the cross section data (b) The determination of the geometric properties of the
was obtained. dielectric, o↵set, minimum and maximum thickness.

Figure C.1.: Data acquisiton for the cross sections of the printed filaments.

(a) Top down image highlighting how the thickness data (b) Top down image for the determination of the o↵set of
of the four layers was obtained. the dielectric.

Figure C.2.: Data acquisiton for the top down images of the printed filaments.

101
C. Printing Process Data

Sensor Core [µm] Shell 1 Shell 3 Dielectric Dielectric Dielectric


[µm] [µm] small [µm] large [µm] average
[µm]
1 231 567 1565 102 235
2 245 584 1579 128 211
3 256 618 1596 131 234
5 182 610 1596 144 289
6 183 568 1563 139 252 195.5
7 179 600 1630 150 260
8 253 631 1544 138 259 198.5
9 269 591 1509 119 231 175
11 334 651 1590 99 202
12 338 672 1503 79 246
13 328 668 1522 99 242
14 335 656 1510 95 261 178
15 331 680 1456 98 242
17 345 679 1468 75 235
23 361 673 1474 106 213
24 396 730 119 181
25 434 750 1573 116 218 167
26 425 767 1578 126 222
27 428 756 124 212
29 397 770 1567 146 217
30 431 807 1627 156 230
31 404 740 1564 106 232
32 429 783 1573 149 211 180
33 435 753 1523 118 214
34 408 709 1540 118 192

Table C.3.: Top down radii values for the filaments that showed no damage under the microscope. In
bold are the sensors that worked and were characterized.

102
D. Readout Electronics
D.1. Additional functions in the arduino file “Static readout half
signal for linear fit.ino”

void CheckforSerialMessage() {
if(Serial.available() > 0) {
// char in order to contain the message that is sent over serial.
char message;

while (Serial.available() > 0) {


message = char(Serial.read());
// Serial.println(message);
delay(2);

// Case 'f'
// "Start Sending the plotting data"

if (message == 'f') {
SendDataPoint = true;
}
// Case 'g'
// "Stop Sending the plotting data"

if (message == 'g') {
SendDataPoint = false;
}
}
}
}

// Functions for data acquisition


// #
void Measure Time response(const int Signal,int Pin, int * TimeSignal, int * SensorResponse, ...

const int buffersize) {

digitalWrite(Signal, HIGH);
unsigned long previousMillis = micros();

for (int i = 0; i < buffersize; ++i) {


unsigned long currentMillis = micros();
TimeSignal[i] = (currentMillis previousMillis);
SensorResponse[i] = analogRead(Pin);
}

digitalWrite(Signal, LOW);
}

// #
// Functions for data processing

void IndexDecayStartandEnd() {

int maximum1 = 0;
IndexDecayStart1 = 0;
IndexDecayEnd1 = buffersize 1;

for (int i = 0; i < buffersize; ++i) {

103
D. Readout Electronics

if (maximum1 < SensorResponseR1[i]) {


maximum1 = SensorResponseR1[i];
IndexDecayStart1 = i + 2;
}

if (SensorResponseR1[i] < maximum1 * 0.135) {

IndexDecayEnd1 = i;
IndexDecayStart1 = IndexDecayEnd1/2;
break;
}
}
}

void TakeLogarithmOfSignal() {

// take the logarithm of the sensor signal


for (int i = IndexDecayStart1; i < IndexDecayEnd1; ++i) {

LogarithmOfSignal1[i IndexDecayStart1] = log(double(SensorResponseR1[i]));

}
}

double calcDecayTime1() {

IndexDecayStartandEnd();
TakeLogarithmOfSignal();

for (int i = IndexDecayStart1; i < IndexDecayEnd1; ++i) {


TimeSignalhilf[i IndexDecayStart1] = TimeSignalR1[i];
}

LinearRegression lr(TimeSignalhilf,LogarithmOfSignal1,IndexDecayEnd1 IndexDecayStart1);


a1 = lr.getA(); b1 = lr.getB(); Rsquare = lr.getCoefDeterm();
return ( 1.0 / b1);

// #
// Functions for data logging

void DoDataLogging(String name, double * timer, double * data,...


int length,int accuracy) {

Serial.println(name);
Serial.println("Time, Signal");

for(int i = 0; i < length; i++) {

Serial.print(timer[i],accuracy);
Serial.print(",");
Serial.print(data[i],accuracy);
Serial.print(",");
Serial.print("\n");

}
Serial.println("EndDataSet");
Serial.flush();
}

void DoDataLogging(String name, int * timer, int * data,...


int length,int accuracy) {

Serial.println(name);
Serial.println("Time, Signal");

for(int i = 0; i < length; i++) {


Serial.print(timer[i],accuracy);
Serial.print(",");

104
D.2. File describtion for the readout version with a multiplexer and two resistors

Serial.print(data[i],accuracy);
Serial.print(",");
Serial.print("\n"); }
Serial.println("EndDataSet");
Serial.flush();
}
}

// Functions for data averaging


// #

void AddDataPoint(double tau) {

sum weights += 1;
sum tau += tau;
sum intersects += a1;

void CalculateAverage() {

average tau = sum tau/sum weights;


average intersects = sum intersects/sum weights;

}
// functions for capacitance and reistance calculation
// #

void calcCapacitance() {

Cs = average tau/(Rs+R1);

// at the beginning of the step response, the voltage is equal


// to a voltage divider of the sensor resistance
// and the measurement resistance

void calcResistance() {

// Serial.println(average intersects);
// Serial.println((double(1024)/exp(average intersects)));

Rs = ((double(1024)/exp(average intersects)) 1)*R1;

D.2. File describtion for the readout version with a multiplexer and
two resistors
All files can be found in the folder: Electronics and Software/Sensor Readout Multiplexer and two
resistors. There are four subfolders, which correspond to an iterative development process of the software.
Not all functions of the GUI’s were implemented and tested. The files should really only serve as a starting
point for new implementations:
• 1 0 Sensor Readout two resistor - This folder contains a very quick implementation for the arduino,
which acquires the decaying voltage signals, linearizes them, calculates the decay times and capac-
itance and resistance via equation 8.16 and 8.17. These values are then transmitted via serial and
the original time signal, the linear fit, decay times, capacitance and resitance change over time are
visualized in a Python GUI (GUI visualization decay times fit resistance capacitance.py)
• 1 1 Sensor Readout Object oriented version - This folder contains a C++ library that does ev-
erything like the script above just in an object oriented manner. Furthermore, it contains a Cal-
ibration.h class that allows to store calibration values for capacitance, resistance and decay time
for certain strains on the microcontroller. However, for this version no GUI was implemented to
actually acquire the calibration values.

105
D. Readout Electronics

• 2 Sensor Readout with most of processing on arduino - These scripts and GUI allow calibration
values (capacitance, resistance and decay times) to acquired for di↵erent strains and stored on the
arduino. However, the internal memory of the arduino was too small in order to store all calibration
values. Yet, it would be sufficient for capacitance values only. This script may serve as a starting
point for a future calibration implementation.
• 3 Sensor Readout Partially object oriented strain calculation occurs on the computer - Here no
calibration data is stored on the arduino. The arduino script has two modes of operation:
– DECAYTIMEMONITORING here the decaying signal and the linear fit are transmitted via
serial and they can be visualized by a GUI (1 GUI for Monitoring of decay.py)
– CAPACITANCE MONITORING - here the capacitance and resistance are transmitted to the
computer and can be visualized by a GUI (2 GUI Readout for static analysis.py) - Important:
This python script can acquire and store calibration values as csv files on the computer and
reload them.

106
E. Sensor charaterization data
E.1. Functionality assessment sensors

(a) Raw Signal for 0 and 50 % strain. (b) Linearization of raw signal and linear fit

Figure E.1.: Sensor 3 Raw Signal and Linearization

(a) Raw Signal for 0 and 50 % strain. (b) Linearization of raw signal and linear fit

Figure E.2.: Sensor 6 Raw Signal and Linearization

107
E. Sensor charaterization data

(a) Raw Signal for 0 and 50 % strain. (b) Linearization of raw signal and linear fit

Figure E.3.: Sensor 8 Raw Signal and Linearization

(a) Raw Signal for 0 and 50 % strain. (b) Linearization of raw signal and linear fit

Figure E.4.: Sensor 11 Raw Signal and Linearization

(a) Raw Signal for 0 and 50 % strain. (b) Linearization of raw signal and linear fit

Figure E.5.: Sensor 13 Raw Signal and Linearization

108
E.1. Functionality assessment sensors

(a) Raw Signal for 0 and 50 % strain. (b) Linearization of raw signal and linear fit

Figure E.6.: Sensor 14 Raw Signal and Linearization

(a) Raw Signal for 0 and 50 % strain. (b) Linearization of raw signal and linear fit

Figure E.7.: Sensor 23 Raw Signal and Linearization

(a) Raw Signal for 0 and 50 % strain. (b) Linearization of raw signal and linear fit

Figure E.8.: Sensor 25 Raw Signal and Linearization

109
E. Sensor charaterization data

(a) Raw Signal for 0 and 50 % strain. (b) Linearization of raw signal and linear fit

Figure E.9.: Sensor 27 Raw Signal and Linearization

(a) Raw Signal for 0 and 50 % strain. (b) Linearization of raw signal and linear fit

Figure E.10.: Sensor 30 Raw Signal and Linearization

110
E.2. Static testing

E.1.1. Functionality assessment long strains

(a) Linearized signal of sensor 8 at very large strains (b) Linearized signal of sensor 9 at very large strains

(c) Linearized signal of sensor 25 at very large strains

Figure E.11.: Two decay times at very large strains for almost all sensors

E.2. Static testing

Sensornumber FitMin [% strain] FitMax [% strain]


6 0 200
8 10 120
9 75 200
14 49 200
25 0 200
32 0 150

Table E.1.: The strain range over which the fit for capacitance and resistance was performed for each
sensor.

111
E. Sensor charaterization data

E.3. Dynamic testing

PreelongationPrestrain PeaktoPeak Maximum Velocity Frequency


[mm] [%] [mm] strain [%] [mm/s] [Hz]
20 33.3% 20 66.6 % 20 0.5
20 33.3% 20 66.6 % 10 0.25
20 33.3% 20 66.6 % 5 0.125
20 33.3% 40 100 % 20 0.25
20 33.3% 40 100 % 10 0.125
20 33.3% 80 166.6 % 20 0.125
20 33.3% 80 166.6 % 10 0.0625

Table E.2.: Actuation amplitude and frequency for all di↵erent dynamic tests.

E.3.1. Connection cable strength load/extension diagrams

(a) Load/Extension diagram connector 1 (b) Load/Extension diagram connector 2

(c) Load/Extension diagram connector 3 (d) Load/Extension diagram connector 4

Figure E.12.: Load/Extension diagrams for the mechanical characterization of the connections.

112
E.3. Dynamic testing

(a) Load/Strain for sensor a (b) Load/Strain for sensor b

(c) Load/Strain for sensor c

Figure E.13.: Load/Strain curves for three sensors until failure.

113

You might also like