Novel Approaches to Evaluating and Characterizing Force Sensor Performance at Body-Device Interfaces
by
Megan Hamilton
A thesis submitted in conformity with the requirements for the degree of Master of Health Science in Clinical Engineering
Institute of Biomaterials & Biomedical Engineering University of Toronto
© Copyright by Megan Hamilton 2019
ii
Novel Approaches to Evaluating and Characterizing Force Sensor
Performance at Body-Device Interfaces
Megan Hamilton
Master of Health Science in Clinical Engineering
Institute of Biomaterials & Biomedical Engineering
University of Toronto
2019
Abstract
Force and pressure sensors are used to monitor fit and inform designers, researchers, and
clinicians in various biomedical applications. However, sensor performance at the body-device
interface is not well understood. The objective of this thesis was to develop and apply novel
approaches to evaluating force sensor performance at clinically-relevant body-device interfaces.
This work includes the characterization of existing transducers as well as the development of a
novel sensor for assistive device applications. Clinically-relevant static and dynamic loading
profiles were applied to two commercial sensors and the prototype sensor via a mechanical
testing machine and an apparatus simulating the lower limb. The results of this study provide
recommendations for use and summarize the effects of area of applied load, sensor
configuration, tissue compliance, and calibration methods. This integration of force sensors is a
vital first step towards adaptable, intelligent assistive devices that have the potential to improve
fit, comfort, and performance.
iii
Acknowledgments
This work was made possible due to the support from a number of individuals, to which I am
deeply indebted.
First and foremost, I would like to thank my supervisor, Dr. Jan Andrysek, for his continuous
support and leadership these past few years. For this, I am extremely grateful. I would also like
to express my appreciation to my co-supervisor, Dr. Kamran Behdinan, and the members of my
thesis supervisory committee, Dr. Eric Diller, Dr. Karl Zabjek, and Dr. Kei Masani, for their
insights, guidance, and dedication in steering this project.
I would also like to express my gratitude to Neil Ready from the Prosthetics and Orthotics
Department at Holland Bloorview Kids Rehabilitation Hospital, for his illuminating feedback
and generous support in developing the methodology.
To the members of the PROPEL Lab I had the privilege and pleasure of working with over the
past few years: Dr. Matthew Leineweber, Dr. Arezoo Eshraghi, Rafael, Calvin, Brock, Mark,
Sam S., Rachel, Emerson, Alex, Firdous, and Harry, thank you for the endless support and
endless laughter. I am grateful to have worked with so many bright, caring, and passionate
researchers. Sam W., thank you as well for your contributions to this study.
A special thanks goes out to Dr. Sharvari Dhote of the ARL-MLS Lab for her assistance in the
study development.
Finally, I wish to express my deepest appreciation and heartfelt gratitude to my loved ones – my
friends and family both near and far – the completion of this work would not have been possible
without you. To my parents, Sara, Penny, and Matt: thank you for your unwavering support; it
means more to me than you know.
iv
Table of Contents
Acknowledgments.......................................................................................................................... iii
Table of Contents ........................................................................................................................... iv
List of Tables ................................................................................................................................ vii
List of Figures .............................................................................................................................. viii
List of Abbreviations ..................................................................................................................... xi
List of Foundations and Funding Sources .................................................................................... xii
Chapter 1 ..........................................................................................................................................1
Introduction .................................................................................................................................1
1.1 Thesis Outline ......................................................................................................................1
1.2 Background and Rationale ...................................................................................................1
1.3 Statement of Research Objectives .......................................................................................4
Chapter 2 ..........................................................................................................................................5
Literature Review ........................................................................................................................5
2.1 Sensing in MAT Applications .............................................................................................5
2.1.1 Types of Transducers for Pressure Measurement ....................................................6
2.1.2 Sensor Evaluation Studies......................................................................................10
2.1.3 Sensing Configurations ..........................................................................................11
2.2 Pressure Distributions at the Body-Device Interface .........................................................12
2.2.1 Pressure Distributions During Gait ........................................................................12
2.2.2 Normal and Shear Loading ....................................................................................14
2.3 Additive Manufacturing .....................................................................................................14
2.3.1 3D-Printed Assistive Devices ................................................................................14
2.3.2 3D-Printed Pressure Sensors ..................................................................................15
Chapter 3 ........................................................................................................................................16
Commercial Sensor Evaluation .................................................................................................16
v
3.1 Load Area and Sensor Configuration ................................................................................16
3.1.1 Introduction ............................................................................................................16
3.1.2 Methods..................................................................................................................18
3.1.3 Results ....................................................................................................................22
3.1.4 Discussion and Conclusions ..................................................................................27
3.2 Tissue Compliance .............................................................................................................30
3.2.1 Introduction ............................................................................................................30
3.2.2 Methods..................................................................................................................31
3.2.3 Results ....................................................................................................................33
3.2.4 Discussion and Conclusions ..................................................................................39
3.3 Dynamic Performance .......................................................................................................41
3.3.1 Introduction ............................................................................................................41
3.3.2 Methods..................................................................................................................42
3.3.4 Results ....................................................................................................................46
3.3.5 Discussion and Conclusions ..................................................................................52
Chapter 4 ........................................................................................................................................54
QTC Prototype Sensor Evaluation ............................................................................................54
4.1 Introduction ........................................................................................................................54
4.1.1 Sensor Fabrication .................................................................................................54
4.2 Methods..............................................................................................................................55
4.2.1 Testing Apparatus ..................................................................................................55
4.2.2 Protocol ..................................................................................................................56
4.2.3 Analysis..................................................................................................................56
4.3 Results ................................................................................................................................56
4.3.1 Calibration..............................................................................................................56
4.3.2 Dynamic Hysteresis Testing ..................................................................................58
vi
4.3.3 Dynamic Simulated Gait Testing ...........................................................................63
4.4 Discussion and Conclusions ..............................................................................................65
Chapter 5 ........................................................................................................................................67
Discussion and Conclusions ......................................................................................................67
5.1 Limitations and Future Work .............................................................................................68
5.2 Significance........................................................................................................................70
References ......................................................................................................................................71
Appendices .....................................................................................................................................77
Appendix A: Explaining the Deadband in Hysteresis Data ......................................................77
Appendix B: QTC Prototype Sensor Inter-Sample Repeatability.............................................78
Appendix C: Additional Data ...................................................................................................79
vii
List of Tables
Table 3-1: Commercial sensor model specifications. ................................................................... 18
Table 3-2: Application conditions: Combinations of variables tested .......................................... 20
Table 3-3: Coefficient of variation values by area and configuration. ......................................... 25
Table 3-4: ANOVA results (p-values) from generalized-area calibration data. ........................... 25
Table 3-5: Specifications for three compliant materials tested. .................................................... 31
Table 3-6: Coefficient of variation values for configuration and material for Peratech and
Sensitronics sensors using matched-area, generalized-area and across configuration calibration
methods. ........................................................................................................................................ 37
Table 3-7: ANOVA results (p-values) from the factorial analysis performed on each sensor
model using generalized-area calibration data. ............................................................................. 37
Table 3-8: Hysteresis error and NRMSE from hysteresis test using both matched-area and
generalized-area calibration for both Peratech and Sensitronics sensors. .................................... 49
Table 3-9: ANOVA results (p-values) for both Peratech and Sensitronics sensors from hysteresis
test results...................................................................................................................................... 49
Table 3-10: Gait simulation testing error values for both Peratech and Sensitronics sensors. ..... 51
Table 3-11: ANOVA results (p-values) from gait simulation testing results. .............................. 51
Table 4-1: Coefficient of variation values for 16 application conditions. .................................... 58
Table 4-2: ANOVA results (p-values) using both matched-area and
generalized-area calibration data. ................................................................................................. 58
Table 4-3: Hysteresis error and NRMSE from two and fifteen-second hysteresis tests using both
matched-area and generalized-area calibration methods. ............................................................. 63
Table 4-4: Gait simulation testing error values for QTC prototype sensor. ................................. 65
Table A-1: CV values calculated for each material, configuration and area for both Peratech and
Sensitronics sensors, using both matched-area and generalized-area calibration methods. ......... 79
viii
List of Figures
Figure 2-1: Strain gauge-based transducers on a prosthetic socket [38]. ....................................... 6
Figure 2-2: Commercially available thin-film interfacial pressure sensors (left to right): Peratech
QTC™, Interlink FSR ®, Sensitronics ® FSR, Tactilus Free Form®, and Tekscan FlexiForce®
A301 [31]. ....................................................................................................................................... 7
Figure 2-3: Apparatus used by Pirouzi et al. to evaluate socket fitting using Tekscan F-Scan
pressure measurement arrays covering the residual limb [28]. ....................................................... 8
Figure 2-4: Experimental apparatus used by Jensen et al [46]. .................................................... 11
Figure 2-5: Loading configuration with and without an elastomer puck highlighting the
difference in effective loading area when the area of applied load is larger than the FSR’s sensing
area. ............................................................................................................................................... 12
Figure 2-6: Typical walking gait phases [47]. Stance and swing phases are labelled for one gait
cycle on the highlighted (right) leg. .............................................................................................. 13
Figure 2-7: Typical interface pressure profile at the residual limb-socket interface of a
transtibial amputee during one stride [48]. ................................................................................... 13
Figure 3-1: Peratech (top) and Sensitronics (bottom) sensors. ..................................................... 18
Figure 3-2: Actual setup (left) and labelled schematic (right) showing configuration with both
loading puck and rigid backing. .................................................................................................... 19
Figure 3-3: Characteristic resistance vs. force curve for the Peratech sensor under all 16
application conditions. .................................................................................................................. 23
Figure 3-4: Characteristic resistance vs. force curve for the Sensitronics sensor under all 16
application conditions. .................................................................................................................. 23
Figure 3-5: Interaction profiles for Peratech sensor using matched-area calibration data. .......... 26
Figure 3-6: Interaction profiles for Sensitronics using matched-area calibration data. ................ 26
Figure 3-7: Characteristic resistance vs. force curves from calibration of compliance testing for
Peratech sensor.............................................................................................................................. 35
Figure 3-8: Characteristic resistance vs. force curves from calibration of compliance testing for
Sensitronics sensor. ....................................................................................................................... 36
Figure 3-9: Interaction profiles for Peratech sensor using simplified generalized-area calibration
CV values. ..................................................................................................................................... 38
Figure 3-10: Interaction profiles for Sensitronics sensor using simplified generalized-area
calibration CV values. ................................................................................................................... 38
ix
Figure 3-11: Force vs. time plots for two-second hysteresis tests for Peratech sensor using
matched-area (subplots i and ii) and generalized-area (subplots iii and iv) calibration methods. 47
Figure 3-12: Force vs. time plots for two-second hysteresis tests for Sensitronics sensor using
matched-area (subplots i and ii) and generalized-area (subplots iii and iv) calibration methods. 48
Figure 3-13: Force measured vs. force applied plots for two-second hysteresis tests for Peratech
sensor using specific matched-area (subplots i and ii) and simplified generalized-area (subplots
iii and iv) calibration methods. ..................................................................................................... 48
Figure 3-14: Force measured vs. force applied plots for two-second hysteresis tests for
Sensitronics sensor using specific matched-area (subplots i and ii) and simplified generalized-
area (subplots iii and iv) calibration methods. .............................................................................. 49
Figure 3-15: Force vs. time plots from dynamic gait testing tests for Peratech sensor using
specific matched-area (subplots i and ii) and simplified generalized-area (subplots iii and iv)
calibration methods. ...................................................................................................................... 50
Figure 3-16: Force vs. time plots from dynamic gait testing tests for Sensitronics sensor using
specific matched-area (subplots i and ii) and simplified generalized-area (subplots iii and iv)
calibration methods. ...................................................................................................................... 51
Figure 4-1: Testing apparatus for QTC prototype sensor. ............................................................ 55
Figure 4-2: Characteristic resistance vs. force curves from QTC prototype sensor calibration
using 16 application conditions. ................................................................................................... 57
Figure 4-3: Force vs. time plots for two-second hysteresis tests for QTC prototype sensor using
specific matched-area (subplots i and ii) and simplified generalized-area (subplots iii and iv)
calibration methods. ...................................................................................................................... 59
Figure 4-4: Force measured vs. force applied plots for two-second hysteresis tests for QTC
prototype sensor using specific matched-area (subplots i and ii) and simplified generalized-area
(subplots iii and iv) calibration methods. ...................................................................................... 60
Figure 4-5: Force vs. time plots for fifteen-second hysteresis tests for the QTC prototype sensor
using matched-area calibration method. ....................................................................................... 61
Figure 4-6: Force vs. time plots for fifteen-second hysteresis tests for QTC prototype sensor
using generalized-area calibration method. .................................................................................. 61
Figure 4-7: Force measured vs. force applied plots for fifteen-second hysteresis tests for QTC
prototype sensor using matched-area calibration method. ............................................................ 62
Figure 4-8: Force measured vs. force applied plots for fifteen-second hysteresis tests for the QTC
prototype sensor using generalized-area calibration method. ....................................................... 62
Figure 4-9: Force vs. time plots from dynamic gait testing calculated using matched-area
calibration data. ............................................................................................................................. 64
x
Figure 4-10: Force vs. time plots from dynamic gait testing calculated using generalized-area
calibration data. ............................................................................................................................. 64
Figure A-1: Raw data and fitted curve for calibration data showing no collected data below 3.5
N. ................................................................................................................................................... 77
Figure A-2: Raw resistance, calculated force, and applied force waveforms displaying deadband.
....................................................................................................................................................... 77
Figure A-3: Characteristic resistance vs. force curve for two QTC prototype sensor samples. ... 78
xi
List of Abbreviations
3D Three dimensional
AM Additive manufacturing
ANOVA Analysis of variance
CB Carbon black
CV Coefficient of variation
DMM Digital multimeter
FSR Force-sensing resistor
GA Generalized-area (calibration method)
HE Hysteresis error
MA Matched-area (calibration method)
MAT Mobility assistive technology
NBNP No rigid backing or elastomer puck (configuration)
NBYP No rigid backing with elastomer puck (configuration)
NRMSE Normalized root mean square error
PDMS Polydimethylphenylsiloxane
PET Polyethylene terephthalate
PLA Polylactic acid
QTC Quantum tunneling composite
RMSE Root mean square error
SD Standard deviation
YBNP Rigid backing with no elastomer puck (configuration)
YBYP Rigid backing with elastomer puck (configuration)
xii
List of Foundations and Funding Sources
EMHSeed, University of Toronto
Natural Sciences and Engineering Research Council of Canada (NSERC) Discovery
RGPIN 2018-05046
Queen Elizabeth II/Victoria Noakes Graduate Scholarship in Science and Technology,
University of Toronto
Barbara and Frank Milligan Award, University of Toronto
Bloorview Research Institute, Holland Bloorview Kids Rehabilitation Hospital
1
Chapter 1
Introduction
1.1 Thesis Outline
The following document consists of five chapters. Chapter 1, beginning here, provides the
rationale for pressure sensing in assistive devices and summarizes the limited understanding of
commercial sensor performance at the body-device interface. Next, the study objectives are
explicitly stated, building from the background and rationale presented.
Chapter 2 provides a survey of the literature on the areas most relevant to the research objectives,
including sensing in mobility assistive technology applications, pressure distributions at the
body-device interface, and a brief description of additive manufacturing of assistive devices and
pressure sensors.
Chapter 3 presents the evaluation of two commercial interfacial pressure sensors using the
methods developed. The three sub-sections outline three different sub-studies: 3.1 examines the
effects of load area and sensor configuration using a static loading protocol. This section was
submitted as a manuscript, including introduction, methods, discussion, and conclusion sections.
Section 3.2 builds off 3.1, with the addition of an advanced calibration loading protocol and
study of an additional experimental factor: tissue compliance. Lastly, Section 3.3 examines the
same commercial sensors using a novel dynamic loading protocol.
Chapter 4 presents the evaluation of a QTC prototype sensor under development, using the
methods previously described in Chapter 3. The sensor is compared to the commercial sensors.
To conclude, Chapter 5 summarizes the main findings and discussion points. Contributions to the
field, limitations to the study, and future directions are also described.
1.2 Background and Rationale
Individuals with physical disabilities can lead fulfilling, independent lives owing to the advances
in wearable sensors and mobility assistive technology (MAT), including prosthetics and
orthotics. The prevalence and financial burden of MAT is evident from the statistical reports; in
2
2016, Medicare approved payment for nearly 3.04 million orthotic services and 2.11 million
prosthetic services, totaling expenditures of USD $1.0 billion and $717 million, respectively [1].
The success of MAT devices, including prosthetics and orthotics, is dependent on their function,
fit, and comfort. There are several negative consequences to a poorly fitting MAT device:
discomfort; excessive stresses, skin irritation, breakdown, and ulcers [2]–[4]; compromised
mobility [5]; and the individual’s decision to discontinue device use [5]. In individuals with
limb-loss, the fit of the socket has been cited as the single most important parameter affecting
user satisfaction [5]. It is also understood that the fit of the socket is dependent on the pressure
distribution at the body-device (i.e. socket) interface [5].
Quantifying the pressure distributions at the body-device interface can help inform the design,
prescription, and fitting of patient-customized MAT devices like prosthetics and orthotics.
Current MAT device fittings require a subjective, manual process dependent primarily on the
experience of the clinician. Reducing the subjectivity in fittings has the potential to improve
device fit, comfort, and usability [6]. In individuals with limb-loss, the body-device interface is
affected by both short and long-term changes, such as prosthesis alignment and setup, growth or
atrophy of the residual limb, and medical complications such as skin lesions, rashes, or pressure
sores. Adjustments must be made by certified prosthetists to accommodate for these changes.
MAT devices incorporating pressure sensors can provide continuous monitoring of the body-
device interface, and enable early detection of conditions that may require clinical intervention
[2], [7], [8].
Force and pressure sensors have been used in a variety of biomedical applications at the body-
device interface, however there is limited data regarding sensor performance in such applications
[9]. Commercial force-sensing resistors (FSR) manufacturers state that consistent actuation and a
rigid surface beneath the sensor are required to ensure repeatable measurement results. When the
area of applied is load is larger than the sensor, manufacturers recommend a thin elastomer
slightly smaller than the sensing diameter, also known as a load “puck”, be placed between the
sensor and load applicator to ensure the force is distributed evenly over the sensing area [10].
However, the body-device interface is subjected to dynamically changing actuation, and the
presence of rigid substrates can cause pain and affect the pressure distribution. Additionally,
commercial FSRs have demonstrated several limitations, including accuracy errors and high
3
hysteresis effects [4], as well as sensitivity to calibration methods and loading conditions [9].
Interfacial sensors require a thin and flexible substrate to minimize effects (i.e., abnormal
pressure distributions) in the system, and a simple manufacturing process that enables integration
into clinical devices [11]–[13].
Due to the complex nature of the body-device interface, it is not feasible to match calibration
conditions to experimental use. For example, the area of applied load will fluctuate throughout
use due to both dynamic movements during gait and changes in the residual limb over time.
Technically, manufacturers would recommend re-calibrating with each area and configuration
that is applied to the sensor – a task that is neither practical nor necessarily possible. There is
thus the need for a simplified, generalized-area calibration method, in-contrast to the traditionally
recommended matched-area calibration method.
Current patient-customized MAT devices require manual manufacturing that is neither time nor
cost effective. These devices do not incorporate pressure sensing due to limitations in sensor
accuracy and ease of integration [14], [15]. Additive manufacturing (AM), commonly referred to
as 3D-printing, shows high potential for the quick and low-cost production of customized
devices with integrated sensors; however, the application of 3D-printing in pressure sensor
development requires further investigation [16]–[18]. There is currently no 3D-printed pressure
sensor available for monitoring body-device interfaces in clinical applications, as no sensors
have demonstrated the performance nor the ease of integration required [12], [18]. The
incorporation of 3D-printing into the fabrication of sensors and assistive devices will produce
cost-effective, comfortable, and intelligent interfaces with the potential to revolutionize the
industry.
A pressure sensor designed to be 3D-printed and embedded within MAT is being developed by
the ARL-MLS (Advanced Research Laboratory for Multifunctional Lightweight Structures) and
PROPEL (Paediatrics, Rehabilitation, Orthotics, Prosthetics, Engineering, Locomotion) labs at
the University of Toronto. The device is a quantum tunneling composite (QTC) pressure sensor
in the initial stages of design and development. The current sensing prototype is fabricated using
manual methods, and its performance is yet to be fully explored. Upon validation of this sensor
prototype, the two labs will begin investigating 3D-printing this sensor (outside the scope of this
study).
4
To validate pressure and force measurements at the body-device interface, there is consequently
a need for a standardized, repeatable method to evaluate and characterize pressure sensor
performance under clinically relevant conditions (i.e., simulating MAT interfaces). This
evaluation should examine multiple factors, including area of applied load, sensor configuration,
and calibration method. This evaluation is needed for both sensors that have been
commercialized and those under development.
1.3 Statement of Research Objectives
The overall objective of this thesis was to develop and apply novel approaches to evaluating and
characterizing pressure sensor performance at clinically relevant body-device interfaces.
The primary aims of this work were as follows:
I. Development of a testing protocol that determines the effects of multiple relevant
parameters and simulates clinically relevant body-device interfaces.
i. Static evaluation of the effects of load area, sensor configuration, and tissue
compliance.
ii. Dynamic evaluation of the effects of load area and sensor configuration.
iii. Examination of the effects of two calibration techniques on sensor performance,
including calibrating under matched-area (i.e. specific) and generalized-area (i.e.
simplified) conditions.
II. Evaluation of both commercially available sensors and those under development.
i. Evaluation and comparative analysis across top-performing commercial FSRs.
ii. Evaluation of the QTC prototype sensor.
5
Chapter 2
Literature Review
2.1 Sensing in MAT Applications
Advances in wearable sensors and sensing technology have the potential to transform traditional
MAT devices into smart-sensing, intelligent devices that can detect and respond to changes in
the user’s environment. Sensing can be used in MAT applications to enable real-time
physiological monitoring, advanced diagnostics, and integrated sensory feedback [18]. Many
different sensors have been developed for use in MAT, including force/pressure [12], strain [19],
tactile [20], displacement[21], angular [22], flow [23], electroencephalography (EEG)[24],
temperature and humidity-monitoring sensors [25], biosensors [26] and accelerometers [18],
[24].
Pressure sensing plays a particularly important role in MAT devices. The results can help inform
the design, prescription, and fitting of patient-customized MAT devices like prosthetics and
orthotics. Current MAT device fittings require a subjective, manual process. Reducing the
subjectivity in a manual fitting has the potential to improve device fit, comfort, and usability [6].
Continuous monitoring can also enable early detection of conditions that may require clinical
intervention, including skin lesions and pressure sores arising from abnormal pressure
distributions [2], [7], [8].
A limited number of studies have been performed to attempt to reduce the subjectivity in
prosthetic fitting, with the objective of reducing the time for both the clinician and the patient to
achieve the best fit, as well as improving the comfort for the patient [5], [27]. One study,
performed by Dakhil et al., demonstrated a correlation between the pain score of patients and
recorded pressures at the body-device interface [27]. The researchers developed a simple,
repeatable clinical protocol using pressure sensors that quantified the pressure distribution at the
body-device interface during both standing and walking trials to improve feedback for prosthesis
fitting and design [27]. Another study, performed by Sewell et al., developed a methodology
using an artificial intelligence approach, specifically inverse problem analysis, to determine
pressures that the body-device interface during standing and walking [5]. The approach
6
calculated the interfacial pressures based off pressure sensors on the exterior of the socket, and
showed promising results, however it is still under development.
2.1.1 Types of Transducers for Pressure Measurement
Pressure sensors have been used in a variety of biomedical applications at the body-device
interface, however there is limited data regarding sensor performance in such applications [9].
Applications include, but are not limited to, prosthetic and assistive devices [12], [28],
laparoscopic instruments [29], tourniquets [30], compression garments [31], helmets [32], in-
shoe gait analysis devices [33], and robotics [34]. A successful sensing element could therefore
be used in a broad variety of applications.
A variety of measurement techniques and sensors have been developed to monitor pressures at
body-device interfaces in order to improve fit and comfort for the user [35]. These sensors
include strain gauges, piezoresistive single-point sensors, piezoresistive sensor arrays, and
capacitive sensors [4].
Strain gauges are a commonly used sensing element that provide high sensitivity and accuracy.
Strain gauges feature small patches of material (i.e., metal) that exhibit a change in electrical
resistance in response to an applied load, and are usually housed in bulky cylindrical shells [4].
However, strain gauges were not designed for measurement at the body-device interface in
clinical settings. Due to their bulkiness, they require permanent modification to an MAT device
for integration (i.e., holes drilled into socket wall to support the strain-gauge and ensure the
sensor is flush against the body) [36]. Another limitation for clinical use is that the strain gauges
have relatively high power requirements for operation; batteries are required for short-term use
[37].
Figure 2-1: Strain gauge-based transducers on a prosthetic socket [38].
7
Piezoresistive sensors are often used for interface pressure measurement due to their thin,
flexible construction, adequate sensitivity, and ease of use [31], [39]. These sensors act as
variable resistors in a circuit: the conductivity of the sensor increases when subjected to an
applied load, as the contact between the conductive particles in the sensor’s polymer matrix
increases. There are several commercially available single point piezoresistive pressure sensors,
displayed in Figure 2-2, including the Quantum Tunneling Composite (QTC™) sensor (Peratech
Ltd, Richmond, North Yorkshire, UK); the FSR® (Interlink Electronics Inc, Westlake Village,
CA, USA); the Tactilus Free Form® sensor (Sensor Products Inc., Madison, NJ, USA); and the
FlexiForce® sensor (Tekscan, Boston, MA, USA). These sensors are often referred to as force-
sensitive resistors (FSRs). One limitation of these sensors is that they measure pressures at a
single point only. For pressure-mapping purposes over a greater area, many single-point sensors
are required, and pressures must be extrapolated at areas between sensors.
Figure 2-2: Commercially available thin-film interfacial pressure sensors (left to right): Peratech
QTC™, Interlink FSR ®, Sensitronics ® FSR, Tactilus Free Form®, and Tekscan FlexiForce®
A301 [31].
Piezoresistive sensing arrays have also been developed to attempt to map pressure distributions
over largers areas, including use within prosthetic sockets. They feature a thin, flexible substrate
and high spatial resolution. The two main commercially available piezoresistive pressure sensing
arrays include Rincoe Socket Fitting (RG Rincoe and Associates, Golden, CO, USA) and F-
Socket (Tekscan, Boston, MA, USA), the latter shown in Figure 2-3. Both piezoresistive sensing
arrays and single-point sensors do not require modification for use in sockets or at other body-
device interfaces, making them a much more viable clinical tool compared to strain gauges.
However, these sensors are susceptible to high hysteresis effects [4].
8
Figure 2-3: Apparatus used by Pirouzi et al. to evaluate socket fitting using Tekscan F-Scan
pressure measurement arrays covering the residual limb [28].
Capacitance-based sensors are constructed of a dielectric material inserted between two parallel
conductive surfaces. The capacitance is a function of the overlapping surface area between the
two conductive surfaces, as well as the distance between them [40]. These sensors have high
precision; however, they are often constructed of rigid substrates and are therefore not suitable
for measurement at body-device interfaces [4]. Laszczak et al. are developing a capacitance-
based sensor, designed for monitoring stresses at the residual limb-socket interface of lower-limb
amputees. The sensor measures both pressure and shear stresses and is constructed of a flexible
3D-printed substrate. The sensor has shown promising preliminary results, but significant
development is required before a product is realized for clinical use [12].
Of the sensors described, single-point piezoresistive sensors, or FSRs, are the most applicable for
monitoring pressure at the body-device interface due to their thin, flexible substrate and small
size that can be tailored to many different applications.
2.1.1.1 Thin-Film Sensors for Interface Pressure Measurement
This section describes the sensor composition and operational principles behind three of the top-
performing thin-film interfacial pressure sensors: the Peratech QTC™ , the Sensitronics
ThruMode™ FSR, and the Tekscan FlexiForce®, as determined by studies conducted by Parmar
and Khodasevych [31], [39].
There are several design criteria required for a successful interfacial pressure sensor. First, the
sensor must be constructed on a thin, flexible substrate. A rigid substrate would create stress-
concentrations at sensor edges, especially at anatomically curved areas, creating tension in the
skin and altering the pressure distribution [41]. The sensor should be able to be easily integrated
9
within the prosthetic socket or other MAT interface, to ensure consistent measurements are
obtained. An accuracy and repeatability of 90% is acceptable in clinical applications [31]. Drift,
or gradual change in output independent of applied pressure, should be quantified to enable
compensation in the calibration to maintain 90% accuracy.
The Peratech QTC™ sensor is a pressure sensitive membrane switch. Carbon finger electrodes
are utilized, extending into a silver electrical connections. A patch of PET printed with carbon
and overprinted with quantum tunneling composite (QTC) ink is assembled over finger
electrodes, separated by an adhesive spacer outlining the active sensing area [42]. The quantum
tunneling composite consists of nickel nanoparticles immersed within an elastomeric polymer
matrix. At rest, the polymer acts as an insulator as the metal particles are not in contact with each
other. Upon deformation, the nanoparticles approach each other, causing the electrical resistance
to drop by several orders of magnitude [43]. Quantum tunneling is achieved by precisely
controlling the shape and blending process used with the nanoparticles. Peratech Ltd. has
commercialized several thin film QTC pressure sensors fabricated using conventional methods
(i.e. not AM).
The Sensitronics ThruMode™ FSR is constructed by affixing two identical layers together
separated by a spacer, where each of two layers is constructed by depositing a solid
semiconductive force-sensing resistor (FSR) element on top of a screen printed solid conductive
silver pad [10]. The FSR element consists of a conductive polymer that operates on the principles
of quantum tunneling, as described previously. The spacer creates a gap when the sensor is
unloaded, and a conductive path between layers when loaded When subjected to a load, the
resistance change is foreseeable (an approximately linear relationship between resistance and
magnitude of the applied load).
The FlexiForce® sensor by Tekscan operates on the same principles as the Sensitronics
ThruMode™ FSR, and is constructed of two layers of polyester substrate that each have a
conductive silver layer, overprinted with a layer of pressure-sensitive ink [44].
10
2.1.2 Sensor Evaluation Studies
Various sensor evaluation studies have been performed; however, the studies have significant
limitations. A sensor evaluation study performed by Ferguson-Pell et al. evaluated the
FlexiForce® for low interface pressure applications. This study included static testing, where
loads were applied to the sensors for periods from 1 min to 2 hours using a low friction tilt table
[45]. The pressure range under study was from 0 to 50 mmHg (0 to 6.7 kPa), targeting interface
pressures reported for residual limb bandages and pressure garments [45]. The researchers
concluded the sensor had acceptable drift, repeatability, linearity, and hysteresis; however,
curvature was identified as a limitation of the sensor. The authors identified future study is
needed on sensor performance when the entire sensing area is not loaded [45].
A study performed by Likitlersuang et al. studied the effects of calibration and sensor-surface
condition (i.e., three conditions: no load disc, rigid disc between sensor and skin, and disc
between sensor and load applicator) on FlexiForce® sensors [9]. The findings revealed that
properly calibrating the sensor with conditions mimicking the test conditions, and placement of a
disc between the sensor and the human body can drastically reduce measurement errors, from
23.3 ± 17.6% to 2.9 ± 2.0 %. The researchers recommended the effects of the area of the applied
load be examined in future work [9].
Parmar et al. published a study in 2017 evaluating the performance of 5 low-cost commercially
available sensors on a human-leg-like test apparatus [31]. The applied pressure range was
between 2.7- 12 kPa, simulating compression therapy applications. The authors used a dead-
weight apparatus and applied simplified static and dynamic loading profiles to the sensors. The
results displayed the Peratech QTC™ was the only sensor that achieved accuracy values of 90%
or higher in both static and dynamic tests. Khodasevych et al. published a study using a similar
apparatus and loading profile to Parmar et al., evaluating the effects of curvature on sensor
performance. Again, the authors found the Peratech QTC™ sensor to be the top-performing
sensor [39].
A study performed by Jensen et al. acknowledged that commercial FSR performance is
dependent on area of applied load. To ensure the pressure sensors studied measured force,
independent of area, an epoxy dome was placed on top of the sensing area of the sensor. Figure
2-4 displays the apparatus used. This prototype was designed for measuring external forces from
11
fingertips (e.g., pinching and grasping tasks), therefore it is not ideal for monitoring the interface
of a larger area as the dome tip can interfere with surface pressures.
Figure 2-4: Experimental apparatus used by Jensen et al [46].
Previous sensor evaluation studies have used simplified dynamic loading patterns [9], [31], [39].
For example, the dynamic loading profiles in studies by Parmar et al. and Khodasevych et al.
were defined as 10 cycles of the application of a load for 30 seconds on and 30 seconds off [31],
[39]. This loading pattern was meant to simulate dynamic wear of a prosthesis; however, it is an
overly simplistic model. Most sensor evaluation studies focus on static testing, in which dead-
weights are loaded onto pressure sensors for time periods of 2.5 minutes [9]; 20 minutes,
representing the time period for adjustments and alternations during a clinical visit [35]; 2 hours
[45]; or 8 hours, simulating full-day wear [31], [39]. Thus, there is a need for a dynamic sensor
evaluation protocol simulating MAT wear and use.
2.1.3 Sensing Configurations
Several different factors and configurations affect FSR performance. Loading pucks, as
recommended by manufacturers, ensure the force applied to the sensor is transmitted fully
through the sensing area [10], [44]. Figure 2-5 displays how the effective loading area changes
with the presence of an elastomer puck when the area of applied load is larger than the sensing
area.
12
Figure 2-5: Loading configuration with and without an elastomer puck highlighting the difference
in effective loading area when the area of applied load is larger than the FSR’s sensing area.
Sensor manufacturers also recommend the sensors be mounted on “relatively rigid” surfaces to
prevent sensor bending and improve uniformity of force distribution across the sensor [10], [44].
However, this is not representative of human tissue, and the presence of a rigid surface has the
potential to alter stress distributions at the interface and cause pain to the user [41].
2.2 Pressure Distributions at the Body-Device Interface
2.2.1 Pressure Distributions During Gait
Typical walking gait consists of two phases: stance-phase and swing-phase. Stance phase is
initiated by heel-strike and refers to the period while the foot is in contact with the ground.
Swing phase, initiated by toe off, refers to the period when the foot is not in contact with the
ground and is propelling forward. One full gait cycle is initiated by the heel strike of one foot
and continues until the same foot prepares to strike for the next step. A single gait cycle is also
known as a stride and is displayed in Figure 2-6.
13
Figure 2-6: Typical walking gait phases [47]. Stance and swing phases are labelled for one gait cycle
on the highlighted (right) leg.
The pressure and force distributions and magnitudes vary at the MAT body-device interface
throughout gait. Figure 2-7 displays the typical interface pressure profile at a residual limb-
socket interface of a transtibial amputee during one stride, from a study performed by Dumbleton
et al. [48]. The points displayed on the plot represent the following: 1=heel contact, 2=peak early
stance, 3=low mid stance, 4=peak late stance, 5=toe-off.
Figure 2-7: Typical interface pressure profile at the residual limb-socket interface of a
transtibial amputee during one stride [48].
A pressure range of 5 to 150 kPa represents conditions applied to the residual limb of lower-limb
amputees during gait [8]. This range encompasses conditions applied to the body under many
other MAT applications, including compressive garments with a range of 5 to 12 kPa [31], and
ankle-foot orthotics with a range of 15 to 130 kPa [33].
14
Both force and pressure measurements are used to describe interfacial distributions in the
literature. Pressure, which is calculated by dividing the applied force by the applied area,
provides insight to the concentration of the force on the tissue, which is typically more indicative
of the pain levels experienced by users than simply absolute force [49]. The range of 5 to 150
kPa can then be translated to forces of 0.4 to 11.8 N given a circular area with a diameter of 10
mm (i.e., sensing diameter of common commercial sensors). In this study, applied loads were
presented with force units (i.e., N). This selection ensured the force remained constant as the area
of applied load was subject to change.
2.2.2 Normal and Shear Loading
Both normal and shear stresses are present at the body-device interface in many MAT
applications. The magnitudes of normal stresses are typically higher than shear stresses, with
reported peak pressures at of a transtibial (below-knee) prosthetic socket site being reported at
roughly 90 kPa for normal stresses and below 10 kPa for shear stresses [50]. The ratio of shear
stress to pressure in a trans-tibial socket interface varies according to the progression of stance,
as well as the location of measurement [51]. Several studies have been performed to determine
the effects of both shear and normal forces on tissue [3], [36], [52]. It was concluded that both
shear and normal stresses have roughly the same effects on tissue, and it is important to consider
the resultant, or combined, stress vector [52].
As the QTC prototype sensor is under development, the initial prototype will target uni-
directional sensing. Thus, as the magnitudes of normal stresses are larger than shear stresses, the
sensor will be initially designed to monitor normal stresses.
2.3 Additive Manufacturing
2.3.1 3D-Printed Assistive Devices
Additive manufacturing has been used in many MAT applications, including prosthetics and
orthotics [14]. In the design stage, a 3D-printed below-knee prosthetic socket with variable
compliance has demonstrated a reduction in contact pressure [16]. Nia Technologies, a Canadian
non-profit social enterprise, is developing 3D-printed prosthetics and orthotics for pediatrics in
developing countries [53]. They are in the process of improving their device design, and have
completed clinical trials in Cambodia, Tanzania, and Uganda. Commercial applications of AM in
15
MAT include the “Cyborg Beast,” a low-cost prosthetic hand for children fabricated using AM
methods [54]. Current limitations to implementation of AM for MAT devices include inadequate
material strength and high initial investment costs [14]. Once these limitations are resolved, full
clinical implementation of AM for custom prosthetics and orthotics is anticipated [14].
2.3.2 3D-Printed Pressure Sensors
A recent industry shift towards AM has led to significant advances in 3D-printed sensing
technology [18]. The need exists for a sensor that can be 3D-printed and easily integrated into
MAT devices. Several researchers have developed pressure sensors in which the sensing element
itself, or the sensor frame is 3D-printed. Laszczak et al. developed a capacitance-based sensor
with a 3D-printed frame for monitoring interface pressure in lower-limb amputees [15]. The
authors identified evaluation of sensor compliance with curved surfaces, sensor flexibility, and
formation of sensor matrices as directions for future work. Lin and coworkers designed an
optical pressure -sensing element in a 3D-printed body, and Saari and coworkers designed a 3D-
printed capacitive force sensor consisting of a frame with embedded wires and a thermoplastic
elastomer [55], [56]. These sensors and others have shown promising initial results, however
further sensor development and evaluation is required for clinical use [12], [56], [57].
Until recently, elastomers could not be 3D-printed due to their material properties. Unlike
thermoplastics often used in AM, elastomers cannot be heated and printed into specific shapes.
Recent advances in printing technology have enabled the 3D-printing of silicone and silicone-
nickel composites [58]. These developments will enable the 3D-printing of quantum tunneling
composite sensors. Previous studies have demonstrated QTC sensors have excellent potential for
pressure sensing applications at the body-device interface [31], [39]. Elastomers will provide the
flexible frame required for interfacial pressure sensors, enabling integration at the body-device
interface.
16
Chapter 3
Commercial Sensor Evaluation
3.1 Load Area and Sensor Configuration
This section contains segments of a manuscript entitled “Evaluating the Effects of Load Area and Sensor
Configuration on the Performance of Force Sensors at Simulated Body-Device Interfaces” authored by
Megan Hamilton, Kamran Behdinan, and Jan Andrysek. This manuscript is currently under review for
publication. The limitations and future work are summarized in Chapter 5 to improve the flow of
information and reduce repetition.
3.1.1 Introduction
Force and pressure sensors can be used at the body-device interface to assess the fit of assistive
devices; provide input for control in robotics; and inform designers, researchers, and clinicians
monitoring interfacial pressure distributions. Single-point thin-film force-sensitive resistors
(FSRs) are often used for interface pressure measurement due to their thin, flexible construction,
adequate sensitivity, and ease of use [31]. There are several commercially available single point
FSRs, including the QTC™ sensor (Peratech Ltd, Richmond, North Yorkshire, UK); the
Sensitronics ThruMode™ FSR (Sensitronics, Bow, WA, USA); and the FlexiForce® sensor
(Tekscan, Boston, MA, USA). Such sensors have been used in a variety of biomedical
applications, however, there is limited data regarding sensor performance at the body-device
interface [8], [9], [31], [39], [59], [60]. Applications include, but are not limited to, prosthetic
and assistive devices [12], [28], laparoscopic instruments [29], compression garments [31], in-
shoe gait analysis devices [33], and robotics [34].
The force-resistance relationship of an FSR depends on sensor shape, geometry, and ink
formulation, as well as actuator geometry and rigidity [10]. Commercial sensor manufacturers’
guidelines state the area of applied load should be held constant at an area slightly smaller than
the sensors’ sensing areas [10], [44], [61]. However, this is not representative of conditions at the
body-device interface, where the loading area fluctuates and typically is larger than the sensing
area. When the area of applied is load is larger than the sensor, manufacturers recommend a thin
piece of silicone rubber 20% smaller than the sensing diameter, also known as a load “puck”, be
placed between the sensor and load applicator to ensure the force is distributed evenly over the
17
sensing area [10]. Sensor manufacturers also recommend the sensors be mounted on “relatively
rigid” surfaces to prevent sensor bending and improve uniformity of force distribution across the
sensor. However, this is not representative of the human tissue, and the presence of a rigid
surface has the potential to alter stress distributions at the interface and cause pain to the user
[41].
A limited number of studies have evaluated sensor performance under conditions that simulate
the body-device interface. A study performed by Likitlersuang et al. studied the effects of
calibration and sensor-surface condition (i.e., three conditions: no load disc, rigid disc between
sensor and skin, and disc between sensor and load applicator) on FlexiForce® sensors [9]. The
findings revealed that properly calibrating the sensor with conditions mimicking the test
conditions, and placement of a thin rigid disc between the sensor and the human body can
drastically reduce measurement errors, from 23.3 ± 17.6% to 2.9 ± 2.0 %. The researchers
recommended the effects of the area of the applied load be examined in future work [9]. Parmar
et al. evaluated the performance of five low-cost commercially available sensors on a test
apparatus simulating the human-leg using deadweights for static and simplified dynamic loading
[31]. The Peratech QTC™ and Sensitronics FSR were found to be the top performing sensors
when drift, accuracy, and repeatability were evaluated. Under static testing conditions, the
Peratech sensor realized an average accuracy in pressure measurements ranging from 83.7 to
98.1%, depending on the force level, while the Sensitronics sensor realized an accuracy ranging
from 84.3 to 87.2% [31]. A similar study by Khodasevych et al. then used the same apparatus
and evaluated the effects of curvature on the Peratech QTC™ and Sensitronics FSR. Their main
finding was that matching calibration conditions (i.e., curvature and compliance) to testing
conditions drastically improved accuracy [39]. Neither of these, nor any other studies, were
found to have examined the effects of area of applied load.
The overall objective of this study was to evaluate the performance of commercially available
thin-film FSRs under conditions simulating the body-device interface. Specifically, the objective
was to determine the effects of load area and sensor configuration (i.e., load puck and rigid
backing) on two commercially available thin-film FSRs under conditions simulating the body-
device interface. A sub-objective of the study was to understand the effects of two calibration
techniques on sensor performance, including calibrating under exact and approximated
conditions.
18
3.1.2 Methods
Testing was conducted to determine the effect of area of applied load on sensor accuracy and
repeatability through various loading puck configurations. Loading configurations following
sensor manufacturer recommendations and simulating more clinically-relevant conditions were
used.
3.1.2.1 Sensors
Two commercially available sensor models, the QTC™ SP200-10 sensor (Peratech Ltd,
Richmond, North Yorkshire, UK) and the ThruMode™ FSR (Sensitronics, Bow, WA, USA),
were selected due to their top performance in similar studies evaluating the performance of
sensors at body-device interfaces [31], [39]. Images of both sensors are displayed in Figure 3-1.
Sensor model specifications are outlined in Table 3-1. Both sensor manufacturers note the
reported part-to-part repeatability is dependent upon consistent actuation.
Figure 3-1: Peratech (top) and Sensitronics (bottom) sensors.
Table 3-1: Commercial sensor model specifications.
Parameter QTC™ SP200-10 Half Inch ThruMode™ FSR
Manufacturer Peratech Ltd. Sensitronics Inc.
Sensing Diameter (mm) 10 12.7
Thickness (mm) 0.45 0.43
Claimed Operating Range (N) 0.1 to 20 0.26 to 26a
Single Part Repeatability (%) N/Ab 5
Part-to-Part Repeatability (%) 4.5 15
a Reported as 0.3 – 30 psi, converted to N using sensing area. b Not reported.
19
3.1.2.2 Data Acquisition System
The Instron Bluehill Universal software (Instron, Norwood, MA, USA) was used for collection
of time and force data. The resistance values (output) of the pressure sensors were measured
using a Keithley 2100 6 ½ Digit Multimeter (DMM) (Tektronix, Inc., Beaverton, OR, USA),
collected using the KI-LINK Excel® Add-In (Tektronix, Inc., Beaverton, OR, USA) and
analyzed using MATLAB (The MathWorks, Inc., Natick, MA, USA).
3.1.2.3 Testing Apparatus
A testing apparatus was developed to simulate human soft-tissue at a body-device interface. A
photo of the actual setup and a labelled schematic showing sensor configuration are displayed in
Figure 3-2. A flat 3D-printed base secured to the Instron base platen via interference fit (PLA
White Material; Ultimaker 2 Printer; Ultimaker B.V., Netherlands) was covered with a 2 cm
outer layer of soft translucent silicone (Renew ® Silicone 10, Renew®, Easton, PA, USA). The
selected silicone type has been proven to mimic the damping and load-dispersal behavior of
human soft-tissue [6], [62], [63]. Loads were applied using an Instron 5944 Universal Testing
System with a 100 N load cell (Instron, Norwood, MA, USA). A force range from 0 to 10 N was
selected [64], [65], to be within both sensors’ working range.
Figure 3-2: Actual setup (left) and labelled schematic (right) showing configuration with both
loading puck and rigid backing.
Loading tip attachments were 3D-printed (PLA, printing process same as described above) and
secured to the upper compression platen via interference fit with the following contact area
diameters: 5 mm, 8 mm, 15 mm, and 25 mm. The smallest 5 mm tip was selected to display
20
effects when only a section of the sensing area is loaded. The 8 mm tip was designed to follow
manufacturer’s recommendations, with the diameter of the load applicator being 20% smaller
than the smallest sensor’s sensing diameter (Peratech) to ensure the spacer and adhesive do not
interfere with the applied force [10]. Larger areas simulate clinically relevant conditions where
the area of applied load is larger than the sensor’s sensing area.
To determine the effects of sensor configuration, a loading puck and rigid backing were used as
per sensor manufacturers’ recommendations. A 1.5 mm thick silicone loading puck (Durometer
60 Shore A) with diameter 8 mm was used to ensure all the force was transmitted through the
sensing area. A 0.6 mm thick polyoxymethylene disc with diameter 14.5 mm was used as a rigid
backing to prevent sensor bending.
3.1.2.4 Protocol
3.1.2.4.1 Application Conditions
To determine the effects of using an elastomer loading puck and rigid sensor backing, a full
factorial experiment was performed on both sensor models using a total of 16 application
conditions: each of four loading tip areas under four loading configurations. The loading
configurations are as follows: i) rigid backing with elastomer puck (as seen in Figure 3-2), ii)
rigid backing with no elastomer puck, iii) no rigid backing with elastomer puck, and iv) no rigid
backing or elastomer puck. Table 3-2 displays the 16 application conditions.
Table 3-2: Application conditions: Combinations of variables tested
Application
Condition
Area
(Diameter (mm)) Backing Puck
1 5 NB NP
2 5 NB YP
3 5 YB NP
4 5 YB YP
5 8 NB NP
6 8 NB YP
7 8 YB NP
8 8 YB YP
9 15 NB NP
10 15 NB YP
11 15 YB NP
12 15 YB YP
13 25 NB NP
14 25 NB YP
15 25 YB NP
16 25 YB YP
21
3.1.2.4.2 Sensor Conditioning and Testing
Prior to each experiment, the sensor was conditioned as per commercial manufacturer guidelines
[44]. This involved applying 110% of the maximum test load, or 11 N, to the sensor for 30
seconds, repeated four times.
For each of the 16 application conditions, forces of 2, 4, 6, 8 and 10 N were applied to the sensor
being tested. The order of applied force and application condition was randomized to account for
potential testing bias. Each force was applied to the sensor for 15 seconds, and force and
resistance values were collected at 2 Hz. The load was then removed for 10 seconds between
forces to reset any drift effects. For each force level under the 16 application conditions, the
average resistance was calculated over the final 7.5 seconds of loading to allow the sensor output
to stabilize. The force and average resistance values then provided the basis for the force-
resistance relationship and enabled the calculation of a resistance vs. force curve for the sensor.
These equations were curve fit using a power relationship as per manufacturer guidelines [10],
which represents the characteristic curve for the application condition and can be used to convert
resistance to force. The entire testing protocol was repeated twice on two different sensors per
sensor model (i.e., two different Peratech sensors and two different Sensitronics sensors), for a
total of four replicates per sensor model.
3.1.2.4.3 Analysis
Sensor performance was assessed using the coefficient of variation (CV). A repeatable FSR
output is required to obtain accurate results. The coefficient of variation (CV) is a measure of
sensor repeatability defined as the ratio of the standard deviation (SD) to the mean. It has been
used to evaluate sensor performance [9], [45]. The Sensitronics manufacturer reports a part-to-
part repeatability of 15% and a single part repeatability of 5%, while the Peratech manufacturer
reports only the part-to-part repeatability (CV) of 4.5% [42], [66]. In general, a CV of 10% is
acceptable in clinical applications [31].
An analysis of variance (ANOVA) was used to examine the effects of the various factors
(diameter, backing, puck, and force) on the CV data for each sensor model (i.e., Peratech and
Sensitronics). All main effects, 2-way and 3-way interactions were evaluated with p<0.05
indicating significance. Insignificant effects were then removed from the model, and significant
22
main effects, 2 and 3-way interactions were reported. JMP® Pro 14 software was used for the
statistical analysis (SAS Institute Inc., Cary, NC, USA).
As part of our sub-objective two different calibration methods were used in this study to
understand the effects of matching the calibration settings to the experimental settings. One
method, the matched-area (MA) calibration, is a more accurate, time-consuming method in
which the exact configuration used during testing is matched during the calibration. The second
method, generalized-area (GA) calibration, is a time-saving approach where one configuration
used during calibration is then applied to multiple configurations during the experimental testing.
For the matched-area calibration, the CV represented the variation in average resistance across
the 4 trials for each configuration, calculated at all 5 force levels and each of 16 application
conditions. For the generalized-area calibration, the CV represented the variation in average
resistance across 4 trials for the given area of applied load compared to the standard 8 mm
applicator tip, calculated at all 5 force levels and each of 16 application conditions. The CV will
be the same for both MA and GA calibrations for the 8 mm conditions. An overall across-
configuration CV method was also calculated, representing the variation across all trials (i.e., all
areas, force levels, and replicates) for a given configuration (i.e., NBNP, NBYP, YBNP, YBYP).
While sensor manufacturers recommend calibration conditions that mimic sensor use, this is not
always possible in clinical situations at the body-device interface (i.e., inconsistent actuation and
area of applied load); hence the relevance of the simplified generalized-area calibration
technique. A pair-wise t-test was performed on each sensor model comparing the MA and GA
calibrations.
3.1.3 Results
The force-resistance data and curve fitting results show the characteristic force-resistance curve
for the four replicates of all 16 application conditions, displayed in Figure 3-3 and Figure 3-4 for
the Peratech and Sensitronics sensors, respectively. Subplots are grouped by configuration: i) no
rigid backing or elastomer puck (NBNP), ii) no rigid backing with elastomer puck (NBYP), iii)
rigid backing with no elastomer puck (YBNP), and iv) rigid backing with elastomer puck
(YBYP). Line colours differentiate the area of applied load, as displayed in the legend. Line
styles (i.e., solid, dashed, dotted and dashed-dotted) indicate the four replicates per sensor. No
23
significant differences were observed between the two sensors tested for each sensor model, so
the sensor number variable was not subjected to further analysis.
Figure 3-3: Characteristic resistance vs. force curve for the Peratech sensor under all 16 application
conditions.
Figure 3-4: Characteristic resistance vs. force curve for the Sensitronics sensor under all 16
application conditions.
24
Matched-area calibration CV is visually represented by the variance in lines of a given colour on
each subplot in Figure 3-3 and Figure 3-4 (i.e., variance across replicates given identical
conditions). Generalized-area calibration CV is visually represented by the variance in all four
lines of a given colour (any area) compared to the black lines (standard 8 mm diameter) on each
subplot. Overall, the Peratech sensor displays increased repeatability compared to the
Sensitronics sensor under the application conditions. For both sensors, the presence of a load
puck appears to greatly improve repeatability, which can be observed in the convergence of the
right-most subplots (with loading puck) for both Figure 3-3 and Figure 3-4, compared to the left
two subplots (without loading puck). The greatest variance is observed in the no-puck conditions
with areas larger than the sensing area (15 and 25 mm diameter).
Table 3-3 summarizes the CV for both Peratech and Sensitronics sensors, using both MA and
GA calibration methods. CV values are displayed for each configuration, averaged over the five
force levels. The error is the standard deviation in CV across force levels. A red-yellow-green
colour gradient was applied to the cells indicating where each CV falls within the entire table
range. The findings from the characteristic curves in Fig. 3 and 4 described above are confirmed
in this table. The Peratech sensor generally outperforms the Sensitronics sensor and the no-puck
and large area conditions display the greatest variance. However, this difference is not
statistically significant, due to the outliers in the Peratech data (i.e., large CV for 15YBNP
configuration). A significant difference between MA and GA calibration methods was reported
only for the Sensitronics sensor. Furthermore, the generalized-area calibration method is best
under conditions when the area of applied load is smaller than the sensor (5 and 8 mm), or in the
presence of a load puck.
25
Table 3-3: Coefficient of variation values by area and configuration. Configuration Coefficient of Variation (%)
Tip
(mm)
Config
Peratech
MA
Peratech
GA
Peratech
Across
Config
Sensitronics
MA
Sensitronics
* GA
Sensitronics
Across
Config
5
NBNP
5.1 ± 1.5 8.3 ± 0.9
44.8 ± 16.5
7.6 ± 3.1 16.7 ± 5.1
108.5 ± 7.6 8 3.1 ± 1.8 3.1 ± 1.8 13.9 ± 6.7 13.9 ± 6.7
15 7.2 ± 4.1 34.0 ± 19.6 48.7 ± 8.0 109.9 ± 6.7
25 36.7 ± 6.5 52.8 ± 13.5 23.1 ± 12.3 92.1 ± 10.3
5
NBYP
5.0 ± 2.3 7.8 ± 1.6
7.7 ± 1.1
11.5 ± 2.7 15.2 ± 3.5
16.4 ± 4.5 8 7.9 ± 2.0 7.9 ± 2.0 15.9 ± 1.2 15.9 ± 1.2
15 6.4 ± 1.4 7.3 ± 1.7 17.3 ± 1.2 16.7 ± 0.9
25 6.4 ± 0.9 7.3 ± 1.5 12.9 ± 5.7 15.4 ± 3.4
5
YBNP
6.9 ± 1.5 16.1 ± 4.8
69.9 ± 37.0
14.6 ± 7.2 12.9 ± 5.0
199.3 ± 51.9 8 16.5 ± 9.1 16.5 ± 9.1 9.1 ± 3.1 9.1 ± 3.1
15 28.1 ± 10.0 56.4 ± 20.8 98.9 ± 35.7 172.7 ± 41.0
25 36.5 ± 46.0 50.0 ± 72.2 113.8 ± 19.2 187.5 ± 23.0
5
YBYP
7.3 ± 1.2 9.8 ± 0.7
9.4 ± 0.7
15.5 ± 2.8 20.5 ± 2.9
21.5 ± 5.0 8 5.2 ± 0.5 5.2 ± 0.5 17.4 ± 5.7 17.4 ± 5.7
15 4.8 ± 0.6 5.4 ± 0.4 18.7 ± 8.5 19.5 ± 4.4
25 4.2 ± 0.5 5.7 ± 1.7 15.1 ± 3.2 18.8 ± 2.7 * Indicates Sensitronics generalized-area calibration is significantly different from matched-area calibration (p
< 0.05) based on pair-wise t-test.
Coefficient of variation values averaged over the five force levels. The error is the standard deviation in CV
across the force levels.
Table 3-4: ANOVA results (p-values) from generalized-area calibration data.
Effect Peratech Sensitronics
Puck <0.0001* <0.0001*
Area <0.0001* <0.0001*
Backing <0.0001* <0.0001*
Force 0.00938* 0.774
Area*Puck <0.0001* <0.0001*
Backing*Puck <0.0001* <0.0001*
Puck*Force 0.03977* 0.872
Area*Backing <0.0001* 0.00113*
Area*Force 0.268 0.960
Backing*Force 0.232 0.26
Area*Backing*Puck <0.0001* 0.00113*
Area*Backing*Force 0.448 0.695
Area*Puck*Force 0.203 0.660
Backing*Puck*Force 0.126 0.410
* Indicates statistical significance (p < 0.05).
Table 3-4 displays the p-values (ANOVA results) calculated for the main effects and two and
three-way interactions. Figure 3-5 and Figure 3-6 show the two-way interaction profiles for the
Peratech and Sensitronics sensors, respectively. Each cell contains line segments plotted for the
26
interaction of the row effect with the column effect. Response values predicted by the model are
connected by line segments. Non-parallel line segments indicate potential interactions, and
significant interactions are confirmed with the p-values in Table 3-4.
Figure 3-5: Interaction profiles for Peratech sensor using matched-area calibration data.
Figure 3-6: Interaction profiles for Sensitronics using matched-area calibration data.
The presence of a 3-way interaction indicates that the lower-order interactions including area,
backing, and puck are conditional on the 3-way interaction. One significant 3-way interaction
27
was observed for both sensors (Area*Backing*Puck). The combination of absent puck and
present backing significantly reduces repeatability at larger areas (diameters of 15 and 25 mm).
Under the no-puck conditions, the 25 mm attachment has better repeatability than the 15 mm
attachment. This can be seen in the Area*Backing and Area*Puck interaction profiles for both
sensors. Repeatability is significantly different only at the larger load applicators (15 and 25
mm), where the presence of a backing increases the CV, and therefore decreases repeatability,
and the presence of the puck improves the repeatability. Significant interactions are also
observed between backing and puck, where the presence of the backing improves the CV only
with the presence of the puck as well.
The presence of a puck had the strongest effect on repeatability for both sensors, with a p-value
of <0.0001. Area was significant for both sensors, with the smaller load applicators resulting in
significantly lower CV results. For both sensors, the backing effect was significant, with no
backing resulting in more repeatable results. Force is a significant factor for the Peratech sensor
(p<0.05), but not the Sensitronics sensor (p=0.77). For the Peratech sensor, the repeatability is
significantly increased for the 10 N force (p=0.02), but significantly lower for the 2 N force
(p<0.05).
3.1.4 Discussion and Conclusions
FSRs are often used for interface pressure measurement in biomedical applications due to their
thin, flexible structure, adequate sensitivity, and ease of use. Manufacturers recommend use on
rigid surfaces with consistent actuation to ensure accurate results [10]. However, the body-device
interface features compliant surfaces and dynamic, inconsistent actuation. This study explored
the effects of area of applied load and sensor configuration (elastomer loading puck and thin
rigid backing disc) on the repeatability of two common commercially available FSRs, the
Peratech and the Sensitronics sensors, under conditions simulating the body-device interface.
The work also examined the effects of matching the calibration settings to the experimental
settings.
Overall, the area of applied load and sensor configuration had significant effects on sensor
repeatability. It should be noted that because the area, puck, and backing factors are involved in a
3-way interaction, the main effects of each are also dependent upon the other factors.
28
The presence of a compliant puck had the strongest effect on repeatability for both sensors, with
a clear indication that the puck improves repeatability. The area also had a significant effect,
where in general, the areas larger than the sensing area (15 and 25 mm diameters) resulted in
greater variance. The presence of a backing also resulted in greater variance overall.
Without the presence of a loading puck, the load tips that are larger than the sensor’s sensing
area apply the load inconsistently. The load is transmitted through the surrounding silicone, the
sensor’s adhesive non-sensing edge, and the sensing area. Therefore, the sensor output will
greatly vary when subjected to the same load, depending how the load is being transmitted. This
effect is most significant for the larger areas without loading pucks.
Under the no-puck conditions, the 25 mm attachment has better repeatability than the 15 mm
attachment. Since the area of the 25 mm attachment is significantly larger than the sensor’s area,
when the load is applied the silicone layer around the sensor is displaced downwards, resulting in
load transmission through the sensing area. The 15 mm attachment is only slightly larger than the
sensor’s total area and is more likely to transfer the load directly through the sensor’s adhesive
layer, not the surrounding silicone. The effect of the backing, in combination with the absence of
a puck and the larger area, reduces sensor bending such that the load is then transmitted through
the sensor’s adhesive layer, not the sensing area. Conversely, for the smaller load applicators (5
and 8 mm), the load is entirely transmitted through the sensing area, regardless of backing or
puck presence.
Force is significant as a 2-way interaction with the use of the puck for the Peratech sensor only.
Thus, the Peratech sensor is more sensitive to the force range, with higher repeatability at higher
forces, compared to the Sensitronics sensor’s performance that is more consistent across the
force range.
In true clinical applications, the body-device interface will apply conditions where the area of
applied load is larger than the sensing area. It is also important to note that this area will also
likely be changing dynamically. To enable repeatable results, the use of an elastomer load puck
that transmits the load through the sensing area is recommended. The presence of a rigid backing
is not recommended based on our results, as it may decrease repeatability. Additionally, the
presence of the rigid layer, which is recommended by manufacturers [10], can alter the stress
distributions at the body-device interface.
29
Significant differences were observed between the matched-area and generalized-area calibration
methods for the Sensitronics sensor only. The Peratech sensor is thus less sensitive to effects of
area of applied load. Without a loading puck, these effects would need to be addressed.
However, the difference in CV between the two calibration methods is decreased with the
presence of a loading puck. This is because the puck transmits the load through the puck area,
ensuring all the load is received by the sensing area. The use of a loading puck thus justifies the
generalized-area calibration technique. Furthermore, the generalized-area calibration method is
best under conditions when the area of applied load is smaller than the sensor (diameters of 5 and
8 mm), or in the presence of a load puck.
Generally, the Peratech sensor provided more repeatable measurements than the Sensitronics
sensor under the various configurations. With the optimal configuration (NBYP – without
backing and with puck), the Peratech CV ranged from 5.0 ± 2.3 to 7.9 ± 2.0 % for the matched-
area calibration method, and 7.3 ± 1.5 to 7.8 ± 1.6 % for the generalized-area calibration
methods. Both ranges are within the acceptable threshold of 10% for clinical use [31]. Under the
same configuration, the Sensitronics CV ranged from 11.5 ± 2.7 to 17.3 ± 1.2 % for the matched-
area, and 15.2 ±3.5 to 16.7 ±0.9 % for the generalized-area calibration, both above the 10%
threshold. Overall, the presence of an elastomer load puck can improve repeatability across all
configuration conditions and enable the use of a generalized-area, time-saving calibration
method.
It is worth noting that the CV values reported in this study agree with the accuracy values
reported in the study by Parmar et al [31]. Parmar et. al reported average accuracy in pressure
measurements ranging from 83.7 to 98.1% (equivalent to a CV of 1.9 – 16.3%) for the Peratech
sensor and 84.3 to 87.2% (equivalent to a CV of 12.8 to 15.7%) for the Sensitronics sensor [31].
These values are also relatively close to the manufacturer reported part-to-part repeatability of
5% for the Peratech sensor and 15% for the Sensitronics sensor. While the findings coincide well
with these previous reports, they provide important new information about the effects of
application-relevant factors such as area and calibration and their interactions, which have not
been previously explored. Limitations to the study are included in Section 5.1.
30
3.2 Tissue Compliance
Evaluating the Effects of Tissue Compliance, Load Area and Sensor Configuration on the
Performance of Pressure Sensors at Simulated Body-Device Interfaces
3.2.1 Introduction
The properties of the tissue in contact with an assistive device are important to consider.
Surrounding tissues are responsible for significant load-bearing and softer tissues promote
uniformity of the pressure distribution, lowering peak interface stresses [67]. Tissue compliance,
or stiffness, has been identified by researchers as a parameter that should be studied to evaluate
interfacial pressure sensor performance at the body-device interface [39]. Limited studies have
explored the effects of material compliance on interfacial sensor performance, however the
studies that have been conducted have significant limitations. A study by Khodasevych et al.
evaluated two different materials: soft (i.e. silicone) and rigid (i.e. aluminum) [39]. There is thus
a need to evaluate the effects of a range of soft materials, representative of tissue.
Other studies have simulated soft tissue for a range of biomedical applications. Examples include
a 5 mm thick silicone elastomer [68], a 20 mm thick polymer [69], and a 20 mm thick silicone
elastomer [31], [39]. The apparent hardness of a material is a function of sample thickness and
the product grade (i.e. manufacturer stated hardness). Apparent or measured hardness increases
inversely with thickness, therefore thicker samples will appear more soft than thinner samples of
the same material [70]. To understand the effects of stiffness on sensor performance, three
different soft materials were evaluated in this section.
Section 3.1 Load Area and Sensor Configuration above included the evaluation of the effects of
load area and sensor configuration (i.e., presence of a rigid backing and load puck) on the
performance of two commercially available sensors. In this section, an additional factor, tissue
compliance, was included in the evaluation. Three different silicone samples of varying hardness
were evaluated.
Specifically, the objective of this section was to determine the effects of tissue compliance, load
area, and sensor configuration (i.e., load puck and rigid backing) on two commercially available
thin-film FSRs under conditions simulating the body-device interface. The effects of two
calibration techniques on sensor performance will be evaluated as a sub-objective.
31
3.2.2 Methods
The sensors and testing apparatus used in Section 3.1 Load Area and Sensor Configuration were
repeated for the testing in this section. The data acquisition system was modified slightly, using a
a Keithley 6500 6 ½ Digit Multimeter (DMM) (Tektronix, Inc., Beaverton, OR, USA) and
Tektronix’s proprietary KickStart software (Tektronix, Inc., Beaverton, OR, USA). Time and
force data were collected using the Instron Bluehill Universal software. All data was collected at
500 Hz. Resistance, force, and time data were analyzed using MATLAB v19 (The MathWorks,
Inc., Natick, MA, USA).
3.2.2.1 Testing Apparatus
The testing apparatus used in Section 3.1 Load Area and Sensor Configuration was repeated for
this section. However, instead of using only one material beneath the sensor simulating tissue,
three different materials were used. The original sample, referred to as the “Extra Soft” material
was selected in addition to two firmer materials. The three different materials evaluated are
described in Table 3-5.
Table 3-5: Specifications for three compliant materials tested.
Identifier Thickness Shore Hardness Product Grade Material Manufacturer
Extra Soft 20 mm 10A Translucent Silicone Renew®
Soft 10A 6.5 mm 10A Silicone McMaster-Carr
Soft 20A 6.5 mm 20A Silicone McMaster-Carr
3.2.2.2 Protocol
3.2.2.2.1 Application Conditions
A total of 48 application conditions were tested. Each of the 16 application conditions listed in
Table 3-2 in Section 3.1 Load Area and Sensor Configuration were repeated with the three
different base materials: Extra Soft, Soft 10A, and Soft 20A. The order of application conditions
selected for testing was randomized to reduce possible testing bias.
3.2.2.2.2 Sensor Conditioning and Testing
Prior to testing each sensor, manufacturer guidelines for sensor conditioning were followed [44],
in which 110% of the maximum test load (11 N) was applied to the sensor for 30 seconds, and
then removed for 30 seconds. This cycle was repeated four times. No data was collected for this
period as the testing was performed solely to condition the sensor before testing.
32
For each of the 48 application conditions, a force sweep from 0 to 10 N was applied to the sensor
at a loading rate of 0.67 N/s (i.e. loading duration of 15 seconds). This force sweep was repeated
three times. The corresponding resistance and force data were curve-fit using an exponential
relationship on MATLAB to characterize the sensors force-resistance curve for each trial [64].
This characteristic curve was then compared across trials and application conditions to study
sensor repeatability.
3.2.2.2.3 Analysis
Sensor performance was assessed using the coefficient of variation (CV) as a measure of sensor
repeatability, as in Section 3.1 Load Area and Sensor Configuration. As stated in the previous
chapter, the Sensitronics manufacturer reports a part-to-part repeatability of 15% and a single
part repeatability of 5%, while the Peratech manufacturer reports only the part-to-part
repeatability (CV) of 4.5% [42], [66]. A CV of 10% is typically acceptable in clinical
applications [31].
An analysis of variance (ANOVA) was used to examine the effects of the various factors
(diameter, backing, puck, and material) on the CV data for each sensor model (i.e., Peratech and
Sensitronics). The specific protocol was described in Section 3.1 Load Area and Sensor
Configuration.
Again, a sub-objective was related to understanding the effects of matching the calibration
settings to the experimental settings. A time-demanding matched-area calibration and a time-
saving, clinically feasible generalized-area calibration method, defined in Section 3.1 Load Area
and Sensor Configuration, were both used. An overall across-configuration CV method was also
calculated, representing the variation across all trials (i.e., all areas and replicates) for a given
configuration (i.e., NBNP, NBYP, YBNP, YBYP with each material). While sensor
manufacturers recommend calibration conditions that mimic sensor use, this is not always
possible in clinical situations at the body-device interface (i.e., inconsistent actuation and area of
applied load); hence the relevance of the generalized-area calibration technique. A pair-wise t-
test was performed on each sensor model comparing the matched-area and generalized-area
calibrations.
33
3.2.3 Results
Figure 3-7 and Figure 3-8 show the characteristic force-resistance curves for the three replicates
of all 48 application conditions for the Peratech and Sensitronics sensors, respectively.
Each figure contains three groupings of subplots for the I) Extra Soft, II) Soft 10A, and III) Soft
20A materials. Within these groupings, subplots are clustered by configuration: i) no rigid
backing or elastomer puck (NBNP), ii) no rigid backing with elastomer puck (NBYP), iii) rigid
backing with no elastomer puck (YBNP), and iv) rigid backing with elastomer puck (YBYP).
Line colours and styles differentiate the area of applied load and trial number, as displayed in the
legend. It should be noted that the y-axes for Figure 3-7 and Figure 3-8 differ, due to differences
in sensor resistance output.
Table 3-6 displays the average coefficient of variation values calculated for each configuration
(I.e., NBNP, NBYP, YBNP, YBYP) and material combination, averaged over the four areas,
calculated for both sensors using both matched-area and generalized-area calibration methods.
Again, matched-area calibration is visualized by the variance in lines of a given colour on a
subplot, generalized-area calibration is visualized by the variance in lines of any given colour
compared with the black lines (8 mm tip) on the same subplot, and the overall CV is visualized
by the variance in all the lines of a given subplot. A red-yellow-green colour gradient was
applied to the cells indicating where each CV falls within the entire table range. The table also
displays the overall across-configuration CV calculated over the four areas and all trials for each
given configuration. Table A-1 in Appendix B displays all data unaveraged from Table 3-6,
highlighting the CV for each area under each application condition.
Overall, the Peratech sensor is significantly more repeatable than the Sensitronics sensor (i.e.,
lower CV), and the matched-area calibration method is significantly more repeatable than the
generalized-area calibration method. The Peratech sensor generalized-area calibration values are
under the 10% desired threshold for all configurations with a puck, regardless of material or
calibration method. For the Sensitronics sensor, CV values averaged 9.4 ± 2.1% for the matched-
area calibration under all configurations, and 11.3 ± 2.1 % for with-puck configurations using the
generalized-area calibration method. The across-configuration CV values approach 100% for no-
puck configurations, however the values remain below 20% for puck conditions.
34
Table 3-7 displays the results (p-values) from the ANOVA test performed on each sensor’s
generalized-area calibration data calculated for the main effects, and 2 and 3-way interactions.
Figure 3-9 and Figure 3-10 display the interaction profiles for the Peratech and Sensitronics
sensors, respectively, using the generalized-area calibration data. As in the previous section, each
cell contains line segments plotted for the interaction of the row effect with the column effect.
Response values predicted by the model are connected by line segments. Non-parallel line
segments indicate potential interactions, and significant interactions are confirmed with the p-
values in Table 3-7. The generalized-area calibration data was selected as it better represents
sensor performance in clinical use, compared to the matched-area calibration data.
35
Figure 3-7: Characteristic resistance vs. force curves from calibration of compliance testing for
Peratech sensor.
36
Figure 3-8: Characteristic resistance vs. force curves from calibration of compliance testing for
Sensitronics sensor.
37
Table 3-6: Coefficient of variation values for configuration and material for Peratech and
Sensitronics sensors using matched-area, generalized-area and across configuration calibration
methods.
Material Config
Average Coefficient of Variation (%)
Peratech
MA
Peratech
GA
Peratech
Across
Config
Sensitronics
MA
Sensitronics
GA
Sensitronics
Across
Config
Extra
Soft
NBNP 2.9 ± 2.3 27.3 ± 24.0 51.9 ± 27.4 10.4 ± 2.7 37.9 ± 32.6 98.6 ± 55.8
NBYP 0.7 ± 0.2 1.8 ± 0.9 3.7 ± 2.4 11.2 ± 3.0 15.6 ± 5.6 18.4 ± 4.3
YBNP 2.6 ± 2.4 27.4 ± 23.4 55.9 ± 33.3 12.7 ± 6.7 37.0 ± 27.8 107.1 ± 47.3
YBYP 1.0 ± 0.4 3.0 ± 2.0 6.6 ± 1.1 8.1 ± 1.6 10.8 ± 3.1 14.1 ± 5.9
Soft
10A
NBNP 9.0 ± 9.5 25.6 ± 22.5 59.0 ± 65.3 7.5 ± 1.2 42.1 ± 37.7 91.6 ± 38.5
NBYP 0.8 ± 0.2 2.8 ± 1.9 5.1 ± 1.1 8.4 ± 1.8 10.8 ± 3.0 14.5 ± 8.2
YBNP 5.9 ± 6.1 31.2 ± 27.0 84.1 ± 53.0 11.1 ± 2.6 35.9 ± 29.2 104.5 ± 34.6
YBYP 1.6 ± 0.8 4.4 ± 3.4 8.9 ± 0.6 6.4 ± 2.5 8.8 ± 2.5 13.5 ± 8.5
Soft
20A
NBNP 7.2 ± 9.6 25.5 ± 22.8 68.6 ± 62.9 11.1 ± 10.8 18.6 ± 11.8 48.4 ± 59.8
NBYP 1.0 ± 0.3 2.3 ± 1.7 4.6 ± 0.8 7.5 ± 0.8 9.9 ± 2.2 14.5 ± 8.2
YBNP 4.9 ± 5.3 30.6 ± 28.4 77.2 ± 40.9 11.6 ± 11.8 16.5 ± 9.1 56.3 ± 79.5
YBYP 1.4 ± 0.3 3.4 ± 1.5 6.2 ± 1.2 6.8 ± 2.2 12.0 ± 3.7 14.3 ± 7.9
Note: Peratech sensor significantly more repeatable than Sensitronics sensor.
Note: Matched-area calibration significantly more repeatable than generalized-area calibration.
Note: No significant 2-way interaction between sensor and calibration, or 3-way interaction between sensor,
calibration and compliance.
Table 3-7: ANOVA results (p-values) from the factorial analysis performed on each sensor model
using generalized-area calibration data.
Effect Peratech Sensitronics
Area 0.0005* 0.0015*
Backing 0.7501 0.4152
Puck <.0001* 0.0003*
Compliance 0.7609 0.0053*
Area*Backing 0.3363 0.3266
Area*Puck 0.0005* 0.0016*
Area*Compliance 0.7585 0.0488*
Backing*Puck 0.7919 0.5761
Backing*Compliance 0.5966 0.6755
Puck*Compliance 0.9904 0.0061*
Area*Backing*Puck 0.2014 0.3377
Area*Backing*Compliance 0.4266 0.8148
Area*Puck*Compliance 0.7767 0.0510
Backing*Puck*Compliance 0.6274 0.6250
* Indicates statistical significance (p < 0.05).
38
Figure 3-9: Interaction profiles for Peratech sensor using simplified generalized-area calibration
CV values.
Figure 3-10: Interaction profiles for Sensitronics sensor using simplified generalized-area
calibration CV values.
Overall, the Peratech sensor exhibited better repeatability (i.e., lower CV) than the Sensitronics
sensor (p < 0.05), regardless of calibration method. For both sensors, the presence of a puck had
the strongest effect on repeatability, with the presence of the puck improving repeatability. This
can be visualized in the decreased variance in the plots on the right-most side (YP conditions),
compared to the plots on the left-most side (NP) conditions in Figure 3-7 and Figure 3-8. The
39
main effect for area and a two-way interaction between area and puck were also significant for
both sensors. Compliance, area*compliance, and puck*compliance were significant for the
Sensitronics sensor only.
The main effect for area and the two-way interaction between area and puck match the results
from Section 3.1 Load Area and Sensor Configuration, where the larger areas exhibit lower
repeatability, and the presence of the puck improves repeatability for the larger areas.
Compliance did not have a significant effect on the Peratech sensor’s performance. However, for
the Sensitronics sensor, the Soft 20A material was significantly more repeatable than the Extra
Soft and Soft 10A materials. There was no significant difference between the Extra Soft and Soft
10A materials for the Sensitronics sensor.
Two-way interactions between i) compliance and area and ii) compliance and puck were
significant for the Sensitronics sensor. The larger areas realized improved repeatability on the
least soft material (Soft 20A), compared to the two softer materials.
The presence of the backing was not deemed significant for either sensor. Neither the main effect
nor any interaction including the backing was statistically significant.
3.2.4 Discussion and Conclusions
This section studied the effects of material compliance, area of applied load, and sensor
configuration (elastomer loading puck and thin rigid backing). Two commercially available
FSRs, the Peratech and Sensitronics sensors were evaluated under conditions simulating the
body-device interface. Secondly, the study examined the effects of i) a matched-area and ii) a
more simplified, generalized-area calibration method. This section built upon Section 3.1 Load
Area and Sensor Configuration, featuring the addition of the material compliance factor and an
updated loading protocol. Commercial FSR manufacturer recommendations for rigid surfaces
beneath the sensor and consistent actuation do not represent the conditions present at the body-
device interface, thus necessitating evaluation of the various biomechanical effects at the
interface.
In clinical applications, the area of applied load and tissue compliance are likely to change
dynamically. Therefore, it is important that a sensor perform repeatably over a range of material
40
compliance values, as well as a range of areas. The generalized-area calibration method in this
study simulates clinical use, where sensors may be calibrated using an initial or standard area,
but throughout use, the area is subject to change.
Overall, the compliance, area of applied load, and sensor configuration had significant effects on
sensor repeatability. As in the previous study, the presence of compliant puck had the strongest
effect on repeatability for both sensors, with the puck presence improving repeatability.
With the use of a load puck, the Peratech sensor was able to consistently provide repeatable
results (less than 10% CV) under testing with all three materials, regardless of calibration
method. Under the generalized-area calibration, the Sensitronics sensor was not able to
consistently achieve repeatable results due to the sensor’s sensitivity to area effects, with a CV
ranging from 8.8 ± 2.5 to 42.1 ± 37.7 % across configurations and materials. It is thus
recommended that the Peratech sensor be used with a load puck for interfacial pressure
monitoring at the body-device interface.
While the compliance or material factor was significant, it is important to note the other
configuration conditions. The backing did not have a significant effect, while the puck did,
overall. This indicates the use of a backing is not warranted to improve repeatability over a range
of compliant surfaces, however the use of a puck is. Limitations to the study are included in
Section 5.1.
41
3.3 Dynamic Performance
Evaluating the Dynamic Performance of Interfacial Pressure Sensors at a Simulated Body-
Device Interface
This manuscript was prepared for submission to IEEE Sensors Journal. The limitations and future work
are summarized in Chapter 5 to improve the flow of information and reduce repetition. An abbreviated
version of the methods is presented and provides references to previous sections to reduce repetition.
3.3.1 Introduction
Pressure sensing in MAT devices can help inform the fitting of patient-customized devices like
prosthetics and orthotics. For example, lower-limb prosthetic fitting and alignment can take
multiple weeks from the first to the final optimized fitting [71]. Each fitting session features both
static (i.e. standing) and dynamic (i.e. walking) weight-bearing assessments, where the
prosthetist relies on their visual perception and fitting experience, as well as amputee feedback to
iteratively refine the fit of the device [71]. While both static and dynamic assessments provide
critical information, dynamic assessments are typically preferable as they provide insight to the
device’s functioning in dynamic, everyday use [72].
However, it is increasingly difficult for a prosthetist to visually perceive the breadth of
information observed in a dynamic walking trial compared to a stationary trial. The integration
of pressure-sensing into the prosthetic fitting process has the potential to reduce the subjectivity
in fitting – enabling prosthetist to base decisions off quantifiable data, not simply their own
observations, experience, and the patient’s feedback. The integration of pressure sensing can also
reduce the number of iterations, or the time, between the initial and final optimized fitting, as
well as improve the overall fit of the device.
There are several commercially available interfacial pressure sensors, including the Quantum
Tunneling Composite (QTC™) sensor (Peratech Ltd, Richmond, North Yorkshire, UK) and
others. Commercially available interfacial pressure sensors have several limitations. The force-
resistance relationship of an FSR depends on sensor shape, geometry, and ink formulation, as
well as actuator geometry and rigidity [10]. Thus, the sensor response is highly dependent on
effects of area of applied load, as displayed in Section 3.1. Sensor manufacturers recommend the
area of applied load should be held constant at an area slightly smaller than the sensors’ sensing
42
areas [10], [44], [61]. Consistent actuation is not representative of conditions at the body-device
interface, where the loading area fluctuates and typically is larger than the sensing area.
Furthermore, through dynamic testing, commercial interfacial pressure sensors have reported
high hysteresis errors [4], [35]. Castellini et al. report that there is a trade-off between the
hysteresis error in a sensor and the static sensitivity, as increased stiffness will alter the
viscoelastic behavior causing hysteresis, but reduced static sensitivity [73].
Most studies that have been performed on commercial sensor evaluation focus on static testing.
The few sensor evaluation studies that have examined dynamic loading used simple dynamic
loading patterns [9], [31], [39]. For example, the dynamic loading profiles in studies by Parmar
et al. and Khodasevych et al. were defined as 10 cycles of the application of a load for 30
seconds on and 30 seconds off [31], [39]. This loading pattern was meant to simulate dynamic
wear of a prosthesis; however, it is an overly simplistic model, considering the average stance
time is approximately one second [74]. Thus, there is a need for a dynamic sensor evaluation
protocol simulating MAT wear and use, as well as a study that examines the effects of area of
applied load on the sensor.
The overall objective of this section was to evaluate the effects of load area and elastomer puck
presence on two commercial FSRs’ dynamic performance. A sub-objective of the study was to
understand the effects of two calibration techniques on sensor performance, including calibrating
under matched-area and simplified, generalized-area conditions.
3.3.2 Methods
Testing was performed the understand the effect of area of applied load on sensor dynamic
performance under two loading configurations: with and without a puck.
3.3.2.1 Sensors
The sensors described in 3.1.2 Methods were used in this section. Peratech reports a hysteresis
error of 8.5% for their sensor, while Sensitronics does not report the value [42], [66].
43
3.3.2.2 Data Acquisition System
The data acquisition system described in 3.2.2 Methods for the compliance testing were used in
this section.
3.3.2.3 Testing Apparatus
An apparatus designed to simulate human tissue was developed for use in a previous study, and
was described in in Section 3.1.2 Methods.
To understand the effects of sensor configuration, a loading puck was used as per sensor
manufacturers’ recommendations under half of the conditions tested. A silicone loading puck
(1.5 mm thickness, 8 mm diameter, and Durometer 60 Shore A hardness) guaranteed the force
was transferred through the sensing area. Previous work (see Section 3.1) indicated the presence
of a rigid backing did not significantly affect sensor performance and was thus not used in this
study.
3.3.3 Protocol
3.3.3.1 Application Conditions
To evaluate the effects of load area and elastomer puck presence on both sensors’ dynamic
performance, a full factorial experiment was conducted using eight application conditions: four
loading tip areas, both with and without an elastomer puck. The order of application conditions
was randomized to minimize potential testing bias.
3.3.3.2 Sensor Conditioning
Prior to testing each sensor, manufacturer guidelines for sensor conditioning were followed [44],
in which 110% of the maximum test load (11 N) was applied to the sensor for 30 seconds, and
then removed for 30 seconds. This cycle was repeated four times.
3.3.3.3 Sensor Calibration
Prior to dynamic testing, a force sweep from 0 to 10 N was applied to the sensor at a loading rate
of 0.67 N/s (i.e. loading duration of 15 s). This force sweep was repeated three times and the
corresponding resistance and force data were curve-fit using an exponential relationship on
MATLAB to characterize the sensors force-resistance curve for a given configuration (i.e. area
44
and puck configuration) [64]. This equation was used to convert resistance output to pressure
values for the subsequent tests.
Two different calibration methods were used: matched-area and generalized-area calibration
techniques. Though sensor manufacturers recommend calibration conditions that imitate sensor
use, this is not always possible at the body-device interface (i.e., inconsistent actuation and area
of applied load); showcasing the importance of the simplified generalized-area calibration
method. Detailed in Section 3.1, the matched-area calibration method is a more accurate, time-
consuming approach that matches calibration settings to experimental settings, and the
generalized-area calibration is a clinically-relevant time-saving approach that applies the
calibration data from one configuration (standard 8 mm area) to multiple experimental
conditions. For this study, both calibration methods were used to convert sensor resistance to
force, enabling a comparison between the result parameters (i.e., NRMSE and HE). Regarding
the matched-area calibration, the calibration equation for each matched-area configuration was
applied to the experimental data. For the generalized-area calibration, the calibration equation
from the 8 mm puck and the given puck configuration was applied to each area.
3.3.3.4 Sensor Testing
3.3.3.4.1 Hysteresis Testing
Hysteresis is the difference in sensor output at the same force when the sensor is being loaded
and unloaded and is commonly used to assess the performance of FSRs [45], [65]. To understand
the sensor’s dynamic performance and identify hysteresis effects, the sensor was loaded from 0
to 10 N and then unloaded to 0 N at a rate of 10 N/s (duration of two seconds). This loading rate
was selected to analyze the hysteresis effects under conditions similar to dynamic loading: one
second each of loading and unloading in the test is comparable to the average stance time of
approximately one second [74].
3.3.3.4.2 Dynamic Simulated Gait Testing
Simulating gait, a dynamic profile, developed as an objective of the study, was applied to the
sensor: loaded to 10 N, held for 1 second, unloaded to 0.5 N, held for 1 second, repeated 10
times. The loaded section of the cycle represents the stance phase of gait, while the unloaded
section represents swing phase. The 10-cycle series represents a 10-step cycle through gait. This
45
test was included as gait trials are integral to fitting sessions for customized MAT specifically, as
well as daily living with MAT.
3.3.3.5 Analysis
Sensor performance was evaluated through analyzing the following measures: accuracy and
hysteresis. Researchers identified these imperative parameters during the evaluation of an
interface force/pressure sensor [39], [41], [45]. Accuracy errors, evaluated in both hysteresis and
gait testing, was calculated using a normalized root-mean squared error (NRMSE). The equation
used to calculate RMSE:
𝑅𝑀𝑆𝐸 =√
∑ (𝐹𝑎𝑝𝑝𝑙𝑖𝑒𝑑 − 𝐹𝑚𝑒𝑎𝑠𝑢𝑟𝑒𝑑)2𝑛
𝑖=1
𝑛
where in this analysis, n is an array of time values beginning at 0 seconds and increasing by
0.002 second increments for the duration of the test. The normalized RMSE, or NRMSE, is then
calculated by dividing by applied force of 10 N and then converting the value to a percentage:
𝑁𝑅𝑀𝑆𝐸 =𝑅𝑀𝑆𝐸
𝐹 ̅∗ 100%
Hysteresis error (HE), evaluated through the hysteresis testing, was calculated by taking the
maximum difference in sensor output (loading versus unloading) for a given force level. The
hysteresis difference, Funloading – Floading, was calculated at each force from 0.5 to 10 N at
increments of 0.1 N. The equation used to calculate HE:
𝐻𝐸 =𝐹𝑢𝑛𝑙𝑜𝑎𝑑𝑖𝑛𝑔 − 𝐹𝑙𝑜𝑎𝑑𝑖𝑛𝑔
𝐹 ̅ ∗ 100%
The hysteresis error was normalized by dividing by the maximum force of 10 N, and then was
converted to a percentage. For each trial, the hysteresis error was calculated at the force with the
greatest hysteresis difference.
An analysis of variance (ANOVA) was used to compare the effects of calibration method and
puck on the NRMSE and HE for each sensor model (i.e., Peratech and Sensitronics). All main
effects, 2-way and 3-way interactions were evaluated with p<0.05 indicating significance.
Insignificant effects were then removed from the model, and significant main effects, 2 and 3-
46
way interactions were reported. JMP® Pro 14 software was used (SAS Institute Inc., Cary, NC,
USA). A paired t-test was performed on each set of results (i.e., NRMSE, HE) to quantify
differences between sensor performance.
As part of the sub-objective, two different calibration methods were used in this study to
determine the effects of calibration. A pair-wise t-test was performed on each sensor model
comparing the matched-area and generalized-area calibrations.
3.3.4 Results
3.3.4.1.1 Dynamic Hysteresis Testing
The force applied vs. time plots for the Peratech and Sensitronics sensors are displayed in Figure
3-11 and Figure 3-12, respectively. The transformed force applied vs. calculated force
representing the hysteresis curves for the Peratech and Sensitronics sensors are displayed in
Figure 3-13 and Figure 3-14, respectively. Subplots are grouped by configuration: left) no rigid
backing or elastomer puck (NBNP), and right) no rigid backing with elastomer puck (NBYP); as
well as calibration method: matched-area (subplots i and ii) and generalized-area calibration
(subplots iii and iv). Line colours distinguish the area of applied load, and line styles distinguish
the trial number, as shown in the legend. NRMSE and HE values for the applications conditions
are displayed in Table 3-8.
In Figure 3-11 and Figure 3-12, the applied force waveform is displayed in green on the plot, as
indicated in the legend. These figures provide a visualization of the sensor’s accuracy in each
configuration. Overall, the Peratech sensor exhibits higher accuracy than the Sensitronics sensor.
The Sensitronics sensor signal exhibits much more noise. The addition of the puck shows a
significant improvement in sensor accuracy, especially in the generalized-area calibration data.
Without a puck, the generalized-area calibration data accuracy reaches roughly 50% for the
larger areas with both sensors. A dead band appears for the Peratech sensor in the NBNP
condition with areas larger than the sensing area (15 and 25 mm), where no force is measured
until approximately 3.5 N. In the matched-area calibration settings, following the dead band, the
data reaches 10 N because each data set was calibrated individually, with a different set of
resistance values corresponding the force values for each configuration. This dead band is further
explained in Appendix A: Explaining the Deadband in Hysteresis Data.
47
Regarding the hysteresis plots in Figure 3-13 and Figure 3-14, the plot for a sensor exhibiting
zero hysteresis would feature a linear segment with a slope of 1, identical for loading and
unloading segments. The difference between loading and unloading curves is thus hysteresis
error. Overall, the hysteresis error is significantly less for Peratech sensor, compared to the
Sensitronics sensor (p=0.01). For both sensors, the addition of the puck appears to reduce
hysteresis error, as the right-most plots (NBYP) display less variance than the left-most plots.
The dead band described above can also be seen in these hysteresis plots. Overall, the effect of
calibration method does not have a significant effect on hysteresis error.
Figure 3-11: Force vs. time plots for two-second hysteresis tests for Peratech sensor using matched-
area (subplots i and ii) and generalized-area (subplots iii and iv) calibration methods.
48
Figure 3-12: Force vs. time plots for two-second hysteresis tests for Sensitronics sensor using
matched-area (subplots i and ii) and generalized-area (subplots iii and iv) calibration methods.
Figure 3-13: Force measured vs. force applied plots for two-second hysteresis tests for Peratech
sensor using specific matched-area (subplots i and ii) and simplified generalized-area (subplots iii
and iv) calibration methods.
49
Figure 3-14: Force measured vs. force applied plots for two-second hysteresis tests for Sensitronics
sensor using specific matched-area (subplots i and ii) and simplified generalized-area (subplots iii
and iv) calibration methods.
Table 3-8: Hysteresis error and NRMSE from hysteresis test using both matched-area and
generalized-area calibration for both Peratech and Sensitronics sensors.
Config Area
Peratech (%) Sensitronics (%)
MA
Calibration
GA
Calibration
MA
Calibration
GA
Calibration
NRMSE HE NRMSE HE NRMSE HE NRMSE HE
NBNP
05 2.6 ± 1.3 8.4 ± 1.7 32.8 ± 0.1 15.9 ± 4.6 5.4 ± 0.5 15.9 ± 1.4 7.8 ± 1.7 20.6 ± 2.8
08 2.0 ± 0.2 8.6 ± 0.4 2.0 ± 0.2 8.6 ± 0.4 5.4 ± 2.1 16.0 ± 4.5 5.4 ± 2.1 16.0 ± 4.5
15 11.1 ± 0.2 23.2 ± 15.3 46.8 ± 0.1 12.8 ± 3.9 9.9 ± 0.2 43.2 ± 8.6 57.3 ± 0.1 39.8 ± 14.6
25 11.0 ± 0.1 12.2 ± 1.6 52.2 ± 0.3 11.5 ± 4.2 11.7 ± 2.1 48.3 ± 7.3 54.9 ± 0.3 38.6 ± 13.2
NBYP
05 4.8 ± 0.4 13.4 ± 0.8 9.2 ± 1.0 15.2 ± 0.6 5.3 ± 1.6 16.8 ± 2.9 9.9 ± 0.4 18.7 ± 2.7
08 1.7 ± 0.4 7.8 ± 1.4 1.7 ± 0.4 7.8 ± 1.4 9.5 ± 0.9 25.4 ± 1.4 9.5 ± 0.9 25.4 ± 1.4
15 2.5 ± 0.2 9.5 ± 0.4 2.8 ± 0.2 8.0 ± 0.1 7.3 ± 1.8 19.8 ± 2.5 9.8 ± 0.7 20.9 ± 1.9
25 2.0 ± 0.3 8.8 ± 0.9 3.0 ± 0.1 6.6 ± 1.0 6.6 ± 2.2 19.0 ± 5.4 8.7 ± 1.1 22.4 ± 3.8
Table 3-9: ANOVA results (p-values) for both Peratech and Sensitronics sensors from hysteresis
test results.
Effect Peratech
NRMSE
Peratech
HE
Sensitronics
NRMSE
Sensitronics
HE
Area <.0001* 0.0241* <.0001* <.0001*
Puck <.0001* 0.025* <.0001* 0.0001*
Calibration <.0001* 0.7189 <.0001* 0.8765
Area*Puck <.0001* 0.0187* <.0001* <.0001*
Area*Calibration <.0001* 0.0382* <.0001* 0.721
Puck*Calibration <.0001* 0.8267 <.0001* 0.5411
* Indicates statistical significance (p < 0.05).
50
For both sensors, the NRMSE is below 10% for the majority of cases, with most exceptions
occurring under conditions without a loading puck at the larger areas. While the HE is under
10% for most conditions for the Peratech sensor, all the HE values for the Sensitronics sensor are
above 10% recommended for use in clinical applications [31].
3.3.4.1.2 Dynamic Simulated Gait Testing
The force-time plots for the gait tests are displayed in Figure 3-15 and Figure 3-16 for the
Peratech and Sensitronics sensors, respectively. Normalized root-mean-squared error values for
both sensors under the different application conditions are displayed in Table 3-10. With the
matched-area calibration method, the NRMSE is below 10% for both sensors, meeting the
desired threshold for accuracy in clinical applications [31]. Using the generalized-area
calibration method with a puck, the CV across the areas averaged 8.8 ± 3.8 % for both sensors.
Without a puck for the generalized-area calibration method, CV values exceed 100% for the
larger areas for both sensors.
Figure 3-15: Force vs. time plots from dynamic gait testing tests for Peratech sensor using specific
matched-area (subplots i and ii) and simplified generalized-area (subplots iii and iv) calibration
methods.
51
Figure 3-16: Force vs. time plots from dynamic gait testing tests for Sensitronics sensor using specific
matched-area (subplots i and ii) and simplified generalized-area (subplots iii and iv) calibration
methods.
Table 3-10: Gait simulation testing error values for both Peratech and Sensitronics sensors.
Configuration Area
NRMSE – Gait (%)
Peratech Sensor Sensitronics Sensor
MA
Calibration
GA
Calibration
MA
Calibration
GA
Calibration
NBNP
05 3.7 ± 1.3 112.9 ± 2.8 8.9 ± 1.6 6.8 ± 0.9
08 6.2 ± 0.3 6.2 ± 0.3 5.0 ± 0.8 5.0 ± 0.8
15 6.6 ± 0.8 118.0 ± 5.9 8.4 ± 0.4 317.9 ± 18.3
25 6.1 ± 0.2 558.3 ± 16.6 5.4 ± 0.6 185.6 ± 6.5
NBYP
05 4.8 ± 0.1 17.2 ± 0.0 4.2 ± 0.8 11.5 ± 1.3
08 4.9 ± 0.6 4.9 ± 0.6 6.8 ± 0.4 6.8 ± 0.4
15 3.7 ± 0.2 6.3 ± 0.3 5.9 ± 0.7 6.3 ± 0.8
25 2.9 ± 0.1 6.8 ± 0.5 7.3 ± 0.1 10.8 ± 1.2
Table 3-11: ANOVA results (p-values) from gait simulation testing results.
Effect Peratech Sensitronics
Area <.0001* <.0001*
Puck <.0001* <.0001*
Calibration <.0001* <.0001*
Area*Puck <.0001* <.0001*
Area*Calibration <.0001* <.0001*
Puck*Calibration <.0001* <.0001*
* Indicates statistical significance (p < 0.05).
52
3.3.5 Discussion and Conclusions
Understanding the dynamic performance of commercial FSRs at body-device interfaces is
critical to support their use in the fitting, prescription, and evaluation of MAT devices. This
study focused on the application of novel dynamic loading, configuration, and calibration
techniques to characterize and evaluate dynamic commercial FSR performance. Previous sensor
evaluation studies failed to examine sensors under dynamic conditions representative of gait at
the body-device interface.
While consistent actuation and rigid mounting is recommended by sensor manufacturers, the
human body exhibits compliant tissue and dynamic actuation [10]. Furthermore, many
researchers have cited hysteresis and accuracy as key requirements of a successful interfacial
sensor, although performance under dynamic conditions is often a limitation [39], [41], [45].
Hysteresis effects, present in dynamic conditions when the sensor is being consistently loaded
and unloaded, reduce the overall accuracy of the sensor. This study determined the hysteresis and
accuracy errors under various conditions representing dynamic loading and used these
parameters to compare sensor models under various configurations.
Overall, the calibration method had a very significant effect on sensor performance, but this is
mitigated with the use of a load puck. The highest errors occurred under conditions with the
larger areas in the absence of a loading puck. This is likely due to the load being transmitted to
surrounding tissue instead of entirely through the sensing area, as described in Section 3.1. The
dead band in the Peratech sensor 15 and 25 NBNP conditions exemplifies this load transmission
complication. Without the puck, the larger area load applicators transmit the load primarily
through the silicone surrounding the sensor, then the sensor’s adhesive layer, and lastly the
sensing area. In these conditions at low forces, the load transmitted through the sensing area was
negligible, and did not produce a change in resistance from the sensor. At a force of
approximately 3.5 N, the surrounding tissue was compressed enough to enable the minimum
activation force to be applied to the sensing area, initiating the curve. The dead band may not
have occurred with the Sensitronics sensor because the sensing area is larger on the Sensitronics
sensor, so more force would be transmitted to it than the Peratech sensor.
Regardless of calibration method, the hysteresis error is significantly lower for the Peratech
sensor (p=0.01) than the Sensitronics sensor. Other improved performance parameters include
53
reduced noise, and higher accuracy overall compared to the Sensitronics sensor. This is likely
due to the chemical formulation of the sensing material, as the Peratech sensor uses a newly
developed QTC™ composite, and the Sensitronics uses a more conventional piezoresistive
material.
Overall, the Peratech sensor provides accurate pressure measurement under both the puck and no
puck configuration conditions for the hysteresis testing using the matched-area calibration
method. With a puck, the hysteresis errors remain below 10% for three of the four areas, and
under 15% for the fourth area. With a puck, this sensor could be used reliably for use at the
body-device interfaces due to the robustness of the sensor’s performance – its reliable
performance regardless of area of applied load and other configuration factors.
Regarding the dynamic gait simulation testing, both sensors exhibit low errors using the
matched-area calibration method. However, the matched-area calibration method is not
necessarily feasible in clinical settings. With the generalized-area calibration method, the use of
a load puck must be used to achieve errors near 10%.
Overall, the dynamic performance of the two commercial sensors is quite similar, however the
Peratech overall exhibited improved performance (i.e. higher accuracy) with lower hysteresis
errors. As in Section 3.1, the authors recommend the use of a load puck under dynamic
conditions to ensure the load is transmitted through the sensor’s sensing area. The use of the
generalized-area calibration method is warranted, given the use of a loading puck.
The NRMSE values reported in this study agree with the accuracy values reported in the study
by Parmar et al [31]. Parmar et. al reported average accuracy in dynamic pressure measurements
ranging from 94.8 to 96.0% (equivalent to an error of 4.0 to 5.2%) for the Peratech sensor and
90.8 to 94.0% (equivalent to an error of 6.0 to 9.2%) for the Sensitronics sensor [31]. These
values are within the range of the errors reported in the 8 mm loading tip applicator conditions
using matched-area calibration. The hysteresis error observed for the Peratech sensor with the 8
mm loading tip applicator conditions using matched-area calibration also agrees with the
manufacturer reported hysteresis error of 8.5% [75]. Though the findings for this sub-set of
configurations agree with previous reports, the information for the range of application-relevant
factors including area and calibration, as well as their interactions, provides new information that
has not been previously explored. Limitations to the study are included in Section 5.1.
54
Chapter 4
QTC Prototype Sensor Evaluation
Development and Evaluation of a Printable Sensor for Pressure-Mapping at Body-Device
Interfaces
4.1 Introduction
A pressure sensor designed to be 3D-printed and embedded within MAT is being developed by
Dr. Kamran Behdinan’s ARL-MLS (Advanced Research Laboratory for Multifunctional
Lightweight Structures) and Dr. Jan Andrysek’s PROPEL (Paediatrics, Rehabilitation, Orthotics,
Prosthetics, Engineering, Locomotion) labs. The device is a quantum tunneling composite
(QTC) pressure sensor in the initial stages of design and development, consisting of a silicone-
nickel composite. The current sensing prototype is fabricated using manual methods, and its
performance is yet to be fully explored. The overall aim of this chapter is to evaluate the
performance of the QTC prototype sensor at a simulated clinically relevant body-device interface
(i.e. lower-limb prosthetic socket), and to compare the sensor to commercially available
interfacial pressure sensors. Upon validation of this sensor prototype, the two labs will begin
investigating 3D-printing this sensor (outside the scope of this thesis).
QTC consists of nickel nanoparticles immersed within an elastomeric polymer matrix. At rest,
the polymer acts as an insulator as the metal particles are not in contact with each other. Upon
deformation, the nanoparticles approach each other, causing the electrical resistance to drop by
several orders of magnitude [43]. Quantum tunneling is achieved by precisely controlling the
shape and blending process used with the nanoparticles. Peratech Ltd. has commercialized
several thin-film QTC pressure sensors that have demonstrated excellent performance [31], [39].
4.1.1 Sensor Fabrication
The QTC prototype sensor was constructed from a spiky spherical nickel (Ni) (T123™, Vale
S.A., Rio de Janeiro, Brazil), carbon black (CB) (ENSACO ® 260 G, Imerys S.A., Paris,
France), and Polydimethylphenylsiloxane (PDMS) (SYLGARD™ 184 Silicone Elastomer, Dow
Chemical Company, Midland, Michigan, United States) composite using a Ni-CB-PDMS mass
ratio of 4:0.14:1 [76]. The sensor chemical composition and development was performed by the
55
ARL-MLS lab and was outside the scope of this thesis. The sample was shaped with a diameter
of 10 mm and a thickness of 1 mm.
4.2 Methods
The methods based off the static compliance testing in Section 3.2 and the dynamic testing in
Section 3.3 were performed on the QTC prototype sensor.
4.2.1 Testing Apparatus
Figure 4-1 displays the testing apparatus used to evaluate the QTC prototype sensor. Thin copper
sheets (0.0025 mm thickness) were cut to match round sensor area and placed on either side of
the sensor. The resistance was measured across the two copper leads using a Keithley 6500
DMM, as described in Section 3.2.
Figure 4-1: Testing apparatus for QTC prototype sensor.
For the calibration, dynamic hysteresis (15-second test), and gait testing, the 16 applications
conditions used match the application conditions outlined in Table 3-2. For the dynamic
hysteresis testing, the two-second tests were performed using 8 application conditions, matching
the configurations used in Section 3.3 without a rigid backing.
56
4.2.2 Protocol
4.2.2.1 Calibration
A force-sweep from 0-10N was applied over 15 s to the sensor under the 16 different application
conditions. See detailed calibration protocol description in Section 3.2.2 Methods.
4.2.2.2 Dynamic Hysteresis Testing
The two-second hysteresis loading protocol outlined in Section 3.3 was repeated for the QTC
prototype sensor. A preliminary analysis of the two-second hysteresis test using the NBNP and
NBYP configurations revealed a slower sensor response time compared to the commercial
sensors. The hysteresis test was then repeated over a fifteen-second period to further understand
the sensor’s transient response, using all four backing and puck configurations (i.e., NBNP,
NBYP, YBNP, YBYP). The fifteen-second period was selected as it is representative of the time
taken during a 10-step gait assessment, standing trial, or rehabilitation exercise [77].
4.2.2.3 Dynamic Simulated Gait Testing
The dynamic gait loading protocol outlined in Section 3.3 was repeated for the QTC prototype
sensor, using all four backing and puck configurations.
4.2.3 Analysis
The analysis in this chapter was performed following the analysis under Section 3.3. Sensor
performance was again evaluated through accuracy and hysteresis parameters, and secondly,
comparisons were performed across calibration methods. However, for the QTC prototype
sensor, a paired t-test was also performed on each set of hysteresis test results (two and fifteen-
second tests) to quantify differences in sensor performance.
4.3 Results
4.3.1 Calibration
The force-resistance plots from the calibration tests are displayed in Figure 4-2. Subplots are
grouped by configuration: i) no rigid backing or elastomer puck (NBNP), ii) no rigid backing
with elastomer puck (NBYP), iii) rigid backing with no elastomer puck (YBNP), and iv) rigid
backing with elastomer puck (YBYP). Line colours and styles differentiate the area of applied
57
load and trial number, as indicated in the legend. As described in Section 3.1.3 Results, the
matched-area CV is represented by the variance in lines of a given colour on each subplot, and
the generalized-area CV is represented by the variance in the lines of a given colour (any one
area) compared to the black lines (standard 8 mm diameter) on each subplot. The across
configuration CV represents the variance in all lines of a given subplot. The coefficient of
variation values are displayed in Table 4-1. The errors displayed represent the standard
deviation.
Figure 4-2: Characteristic resistance vs. force curves from QTC prototype sensor calibration using
16 application conditions.
58
Table 4-1: Coefficient of variation values for 16 application conditions.
Configuration Area Matched-Area
CV (%)
Generalized-Area
CV (%)
Across Configuration
CV (%)
NBNP
05 4.7 ± 6.5 14.9 ± 14.7
19.9 ± 14.7 08 6.3 ± 1.7 6.3 ± 1.7
15 9.9 ± 11.2 16.2 ± 7.7
25 11.0 ± 5.6 13.5 ± 4.1
NBYP
05 8.8 ± 19.9 20.5 ± 18.6
28.6 ± 12.8 08 4.3 ± 3.4 4.3 ± 3.4
15 7.0 ± 8.2 8.0 ± 12.4
25 10.1 ± 3.5 21.5 ± 3.1
YBNP
05 2.8 ± 3.6 11.3 ± 10.3
13.8 ± 11.6 08 4.3 ± 6.2 4.3 ± 6.2
15 13.6 ± 11.9 13.2 ± 10.8
25 3.0 ± 1.5 6.5 ± 5.8
YBYP
05 5.7 ± 7.9 21.1 ± 7.0
20.7 ± 10.6 08 6.1 ± 3.3 6.1 ± 3.3
15 4.9 ± 2.8 25.7 ± 6.1
25 4.7 ± 2.3 16.2 ± 3.7
Table 4-2: ANOVA results (p-values) using both matched-area and
generalized-area calibration data.
Effect QTC Matched-Area QTC Generalized-Area
Backing 0.143 0.962
Area*Puck 0.178 0.511
Area*Backing 0.245 0.347
Area 0.245 0.088
Puck 0.684 0.122
Backing*Puck 0.959 0.180
* Indicates statistical significance (p < 0.05).
Overall, neither the effects of the backing, area, puck, nor the interactions of each combination
were statistically significant (p>0.05).
4.3.2 Dynamic Hysteresis Testing
Figure 4-3 and Figure 4-4 display the force vs. time and measured force vs. applied force plots
for the two-second hysteresis test, respectively. For both figures, subplots are grouped by
configuration and calibration: i) NBNP Matched-Area Calibration, ii) NBYP Matched-Area
Calibration, iii) NBNP Generalized-Area Calibration, and iv) NBYP Generalized-Area
Calibration. The NBNP appears to have slightly higher accuracy than the NBYP condition for
the matched-area calibration data. For the generalized-area calibration data, there does not appear
59
to a difference across configurations, except for the 5 mm diameter tip appears to have high
errors under the NBNP configuration. Hysteresis errors do not appear to be significantly different
across configurations and areas, as the overall distance between loading and unloading curves
(i.e., visual representation of hysteresis error) remains relatively constant.
Figure 4-3: Force vs. time plots for two-second hysteresis tests for QTC prototype sensor using
specific matched-area (subplots i and ii) and simplified generalized-area (subplots iii and iv)
calibration methods.
60
Figure 4-4: Force measured vs. force applied plots for two-second hysteresis tests for QTC
prototype sensor using specific matched-area (subplots i and ii) and simplified generalized-area
(subplots iii and iv) calibration methods.
Figure 4-5 and Figure 4-6 display the force vs. time plots for the fifteen-second hysteresis test
using matched-area and generalized-area calibration methods, respectively. Figure 4-7 and
Figure 4-8 display the measured force vs. applied force for the same data sets as Figure 4-5 and
Figure 4-6. For the figures, subplots are grouped by configuration: i) NBNP, ii) NBYP, iii)
NBNP, and iv) NBYP. As in the other sections, line colour and style represent the load tip area
and trial number, as indicated in the legend.
Table 4-3 displays the CV calculated using both matched-area and generalized-area calibration
methods from both two and fifteen-second tests on the QTC prototype sensor.
61
Figure 4-5: Force vs. time plots for fifteen-second hysteresis tests for the QTC prototype sensor
using the matched-area calibration method.
Figure 4-6: Force vs. time plots for fifteen-second hysteresis tests for the QTC prototype sensor
using the generalized-area calibration method.
62
Figure 4-7: Force measured vs. force applied plots for fifteen-second hysteresis tests for the QTC
prototype sensor using the matched-area calibration method.
Figure 4-8: Force measured vs. force applied plots for fifteen-second hysteresis tests for the QTC
prototype sensor using the generalized-area calibration method.
63
For the fifteen-second hysteresis test data, accuracy (displayed through NRMSE) and hysteresis
errors appear relatively constant across configurations and calibration methods.
Table 4-3: Hysteresis error and NRMSE from two and fifteen-second hysteresis tests using both
matched-area and generalized-area calibration methods.
Config Area
2 s Hysteresis Test 15 s Hysteresis Test
MA
Calibration
GA
Calibration
MA
Calibration
GA
Calibration
NRMSE HE NRMSE HE NRMSE HE NRMSE HE
NBNP
05 13.0 ± 2.7 27.0 ± 5.6 9.4 ± 1.1 24.5 ± 4.3 9.4 ± 1.1 24.5 ± 4.3 8.0 ± 1.0 19.0 ± 4.2
08 14.3 ± 2.3 37.6 ± 7.9 7.1 ± 1.7 16.6 ± 3.7 6.9 ± 1.7 16.3 ± 3.7 6.9 ± 1.7 16.3 ± 3.8
15 12.6 ± 1.7 34.8 ± 3.6 12.9 ± 1.0 31.3 ± 4.1 10.4 ± 0.7 23.6 ± 0.9 18.0 ± 1.5 19.7 ± 1.4
25 12.9 ± 1.5 36.6 ± 2.1 12.8 ± 3.9 30.9 ± 8.4 12.8 ± 3.9 30.9 ± 8.4 11.3 ± 1.3 16.9 ± 7.0
NBYP
05 9.5 ± 0.7 20.5 ± 4.2 9.6 ± 0.8 25.1 ± 1.7 9.6 ± 0.8 25.1 ± 1.7 23.2 ± 1.8 24.6 ± 3.6
08 15.6 ± 2.3 30.6 ± 3.8 8.9 ± 0.2 26.2 ± 1.0 8.1 ± 0.2 23.8 ± 1.0 8.1 ± 0.2 23.8 ± 1.7
15 12.3 ± 0.6 40.0 ± 3.7 9.4 ± 0.8 27.9 ± 0.2 9.4 ± 0.8 27.9 ± 0.2 6.9 ± 0.8 20.8 ± 1.9
25 12.4 ± 0.8 28.9 ± 5.4 16.8 ± 3.1 26.8 ± 5.4 16.8 ± 3.1 26.8 ± 5.4 12.3 ± 0.3 31.5 ± 6.5
YBNP
05 9.2 ± 2.0 26.5 ± 4.8 11.9 ± 0.3 27.7 ± 2.0
08 11.2 ± 2.2 26.1 ± 2.9 11.2 ± 2.0 26.1 ± 2.2
15 12.4 ± 0.9 28.7 ± 2.7 12.0 ± 0.7 26.6 ± 2.9
25 7.8 ± 0.4 22.3 ± 1.2 7.5 ± 0.1 23.3 ± 1.4
NBYP
05 9.4 ± 1.0 23.5 ± 2.1 16.9 ± 2.2 23.9 ± 4.0
08 6.9 ± 0.9 19.9 ± 1.7 6.9 ± 0.8 19.9 ± 1.6
15 11.0 ± 1.4 26.7 ± 1.6 15.8 ± 2.6 16.3 ± 2.4
25 8.6 ± 3.6 25.4 ± 5.1 15.2 ± 0.9 30.1 ± 1.9
Comparing results across the two and fifteen second tests, all errors were lower for the fifteen-
second test compared to the two-second test (p<0.05). Comparing results between calibration
methods, significant differences (p<0.05) were observed between two-second NRMSE
(matched-area resulted in lower errors), and fifteen-second HE (generalized-area resulted in
lower errors). All other parameters were not significantly different across calibration methods.
4.3.3 Dynamic Simulated Gait Testing
Figure 4-9 and Figure 4-10 display the force vs. time profiles from dynamic gait testing
calculated using the matched-area and generalized-area calibration methods, respectively.
Table 4-4 displays the NRMSE for the dynamic simulated gait testing calculated using both
matched-area and generalized-area calibration data. The generalized-area calibration data results
in significantly higher errors than the matched-area calibration, however neither calibration
64
method achieves high accuracy. Overall, none of the four backing-puck configurations realize
consistent errors below 10%, regardless of calibration method.
Figure 4-9: Force vs. time plots from dynamic gait testing calculated using matched-area
calibration data.
Figure 4-10: Force vs. time plots from dynamic gait testing calculated using generalized-area
calibration data.
65
Table 4-4: Gait simulation testing error values for QTC prototype sensor.
Configuration Area
NRMSE – Gait (%)
Matched-Area
Calibration
Generalized-Area*
Calibration
NBNP
05 13.6 ± 1.1 16.4 ± 0.8
08 15.3 ± 4.0 15.3 ± 4.0
15 25.8 ± 1.6 35.6 ± 0.8
25 18.6 ± 1.5 28.9 ± 1.2
NBYP
05 17.1 ± 0.9 6.2 ± 0.9
08 13.0 ± 0.2 13.0 ± 0.2
15 19.5 ± 2.3 20.3 ± 0.9
25 6.7 ± 0.3 15.2 ± 0.9
YBNP
05 26.8 ± 1.1 25.8 ± 0.6
08 15.2 ± 0.7 15.2 ± 0.7
15 17.0 ± 1.0 21.4 ± 1.0
25 18.1 ± 1.5 20.0 ± 1.5
YBYP
05 9.1 ± 2.5 25.9 ± 2.6
08 20.8 ± 1.9 20.8 ± 1.9
15 12.6 ± 0.8 31.3 ± 0.9
25 13.5 ± 0.7 27.7 ± 0.7
*Indicates significantly different from matched-area calibration.
4.4 Discussion and Conclusions
For calibration results, the generalized-area CV and across-configuration CV values provide the
most realistic representation of the sensor’s performance in clinical use. Calibration during
clinical use will likely be brief, and the area of applied load may change dynamically. With
across-configuration CV values ranging from 13.8 ± 11.6 to 28.6 ± 12.8%, the desired 10%
threshold for use in clinical applications was not attained [31]. Further revisions to the sensing
prototype are recommended to achieve repeatable results.
Over the two-second hysteresis testing, the sensor displayed a delayed time response. Firstly, the
peak force measured by the sensor was not aligned with the peak applied force. Secondly, the
sensor’s peak measured response did not reach 10 N under any condition. This lengthy sensor
response time could be attributed to the thickness of the sample (1 mm), compared to Peratech’s
sensing thickness of 8 to 10 µm [42]. The increased thickness creates a greater distance for the
electrons to travel. This delayed response was also displayed though the high error rates in the
two-second hysteresis and gait testing. Both tests were performed at a high loading rate and
frequency that did not provide the sensor with enough time to respond. The improved accuracy
displayed in the fifteen-second hysteresis tests also validate this theory. Under the fifteen-second
66
hysteresis tests, the QTC prototype sensor’s accuracy approaches 10% over most conditions,
indicating its feasibility as a sensing prototype.
The configuration (i.e., presence of a backing and puck) did not have a strong effect on the QTC
prototype sensor’s performance. As the entire sensor is constructed of the compliant sensing
component, the loading puck does not aid in ensuring transmission through the sensing area.
Additionally, the sample was relatively thick and minimal bending occurred throughout testing,
so the presence of a rigid backing also had a negligible effect.
The generalized-area calibration method resulted in a higher normalized root mean squared
errors for both hysteresis and gait tests compared to the matched-area calibration method, as
expected. However, the error rates for both calibration methods exceed the desired 10%
threshold [31].
In general, calibration influenced accuracy; however, it did not affect hysteresis error. This is
expected, as the calibration method will change the equation converting resistance to force but
does not affect the difference in sensor output between loading and unloading curves.
Other studies published on the development of new sensors typically present results that compete
with commercially available products, with accuracy typically above 90% [78]. Future
development is required on the current sensing prototype to achieve competitive performance.
Limitations and future work are described in Section 5.1.
67
Chapter 5
Discussion and Conclusions
The methods developed and applied in this work were used for evaluating and characterizing
pressure sensor performance at clinically relevant body-device interfaces. This study introduced
the evaluation of various clinically relevant factors on sensor performance – including the effects
of load area, sensor configuration, tissue compliance, and calibration method. A dynamic loading
protocol was also developed, simulating gait. The conducted evaluation of the commercial
Peratech and Sensitronics sensors can be used to inform the design of pressure-mapping mobility
assistive technology. The evaluation of the QTC prototype sensor provides a preliminary
analysis of its performance, informing future iterations of the design. The contributions of this
work regard to both methodology and sensor development.
Through the static evaluation of the commercial sensors, the effects of load area, sensor
configuration, and tissue compliance were examined. While the area of applied load and tissue
compliance both have significant effects on sensor repeatability, this can be mitigated using a
load puck. The use of a backing is not warranted for the conditions simulated (i.e. uniform
tissue). Through the dynamic evaluation, the presence of the puck was again proven critical.
Overall, the Peratech sensor provided more repeatable, more accurate measurements (i.e., lower
CV, HE, and NRMSE) than the Sensitronics sensor under the various configurations, for both
static and dynamic loading. The secondary objective, evaluating the effects of calibration using
exact and approximated conditions, indicated that a generalized-area calibration technique can be
used in pair with a load puck, and the Peratech sensor is recommended. The Peratech sensor
proved to have the most robust design, and its performance remained consistent when subjected
to the various configurations.
The evaluation of the novel sensor provided valuable information to guide in designing future
iterations of the sensing prototype. The sensor configuration (i.e. backing and puck) and area of
applied were not significant to sensor performance. This is due to the current prototype’s
composition being entirely a sensing component, unlike commercial sensors with adhesive non-
sensing edges. While the matched-area calibration method had more repeatable and higher
accuracy results, both calibration methods had similar results (exceeding desired 10% error
threshold). While this sensor does not currently compete with the performance of the commercial
68
sensors, the sensing prototype has shown significant potential. The potential to 3D-print and
integrate the QTC prototype sensor will be extremely valuable for MAT applications.
5.1 Limitations and Future Work
Limitations to the study should be noted. Firstly, frictional shear forces were not accounted for,
and are assumed to be minimal given the perpendicular loading. However, shear forces can have
significant effects at the body-device interface in real-life applications. A study by Zhang et. al
concluded that both shear and normal stresses have roughly the same effects on tissue, and it is
important to consider the resultant, or combined, stress vector [52]. Secondly, temperature and
curvature effects, which have been examined in other works [39], [64], were not examined in this
study to allow the research question to prioritize the effects of area of applied load and
configuration.
Improved sensor performance was observed using the dynamic force-ramp loading method
(absence of drift) in Section 3.2 compared to the static loading at multiple force levels (presence
of drift) method used in Section 3.1. This displays the errors introduced by drift. Drift was not
included in either study because it has been studied extensively by other works, however its
contribution to the static loading protocol is important to note.
While this study featured unique calibration and loading protocols, the conditioning methods
used were those recommended by manufacturers. Conditioning, or exercising, is the process of
cyclically loading and unloading the sensor before calibration and actual testing. Several other
studies have used the same conditioning method, as it has become the standard in the field [31],
[39], [64]. Tekscan, the manufacturer of common FlexiForce ® sensors, states that the process of
conditioning helps to lessen the effects of drift and hysteresis [44], however there is limited data
validating such conditioning methods [79]. In a study performed by Hall et al., the sensitivity to
shear loading was eliminated by extended sensor preloading in compression and shear [79]. In
the same study, they were able to compensate for hysteresis errors by including terms dependent
on loading history in the calibration equation [79]. Future work could examine the effects of
various conditioning methods on parameters including accuracy, hysteresis, and drift.
Future work examining dynamic loading should examine various force/pressure ranges to
simulate the pressure distributions at a range of locations within a body-device interface, as well
69
as a range of activities. For example, the forces at both peaks and valleys can be customized, as
well as the loading rates.
It is important to acknowledge the assumption of uniformity in tissue loading present in this
paper. The testing performed featured a uniform layer of silicone simulating tissue at the body-
device interface. In true applications, the unique anatomy of any limb (i.e., presence of bony
prominences) presents many inconsistencies in tissue properties (e.g., varying compliance,
curvature, etc.). While the presence of the thin rigid backing did not have a significant effect in
this work, this would not likely hold true with nonuniform tissue. Future studies could examine
the effects of nonuniformity in tissue on the sensor’s performance.
Additionally, this study focused on the evaluation of single point FSRs. While the information
regarding the pressure at a single point is valued, it is more valuable to map the pressure
distributions over the entire residual limb during gait [48]. Upon validation of the QTC prototype
sensor, future work should examine creation of sensing arrays.
While the performance of the QTC prototype sensor did not quite meet commercial sensor
standards, the sensor showed promising results for an initial sensing prototype. Future iterations
should focus on thinning the QTC sensing layer (i.e. screen printing), as this may improve sensor
response time, as well as improve the integrability within systems. Future work will also be
required to determine and refine sensor electrical integration properties, as the thin copper layer
is appropriate for bench top testing purposes only. Additionally, future work on sensor
development should study the sensor response time and drift effects under clinically relevant
conditions.
Upon completion of benchtop testing validating the sensor’s performance, a clinical pilot study
will be conducted. This will consist of the integration of the sensor into an assistive device,
creating an interface with embedded sensors. A conventional lower-limb prosthetic socket or
orthosis will be modified, and a small clinical pilot study will be conducted, comparing the QTC
prototype sensor with a gold-standard sensor in pressure measurement (e.g., commercial strain
gauge).
70
5.2 Significance
In this work, novel approaches to evaluating force sensor performance at clinically relevant
body-device interfaces were developed and applied. This work includes the characterization of
existing transducers as well as the development of a novel sensor for assistive device
applications. Beyond the scope of this thesis, the overarching goal of this work was to advance
the use of interfacial pressure sensors at the body-device interface to facilitate quantification of
fit of MAT devices, and improve comfort, function, and satisfaction for users. The development
of a novel testing protocol simulating clinically relevant body-device interfaces has the potential
to disrupt current sensor evaluation methods. The novelty of the protocol arose from the dynamic
gait simulation, as well as the study of the effects of area of applied load and sensor
configuration. This method can be used to better understand sensing at the body-device interface,
and aid in developing new sensing technologies.
The evaluation of the QTC prototype sensor will act as a proof-of-concept for further
development of the technology, and 3D-printed assistive devices with embedded sensors. Fully
integrating sensors for measuring biomechanical and physiological aspects of prosthetic and
orthotic function is an essential first step towards fully adaptable intelligent prostheses that can
both detect and respond to the user’s changing physiology, anatomy, functional requirements, as
well as activities and environment.
71
References
[1] “AOPA Fact Sheet,” The American Orthotic & Prosthetic Association, 2016. [Online].
Available: http://www.aopanet.org/media/fact-sheet/. [Accessed: 01-Oct-2017].
[2] S. W. Levy, “Skin problems of the leg amputee.,” Prosthet. Orthot. Int., vol. 4, no. 1, pp.
37–44, 1980.
[3] N. A. Abu Osman, W. D. Spence, S. E. Solomonidis, J. P. Paul, and A. M. Weir,
“Transducers for the determination of the pressure and shear stress distribution at the
stump-socket interface of trans-tibial amputees,” Proc. Inst. Mech. Eng. Part B J. Eng.
Manuf., vol. 224, no. 8, pp. 1239–1250, 2010.
[4] E. A. Al-Fakih, N. A. Abu Osman, and F. R. Mahmad Adikan, “Techniques for interface
stress measurements within prosthetic sockets of transtibial amputees: A review of the
past 50 years of research,” Sensors (Switzerland), vol. 16, no. 7, 2016.
[5] P. Sewell, S. Noroozi, J. Vinney, R. Amali, and S. Andrews, “Static and dynamic pressure
prediction for prosthetic socket fitting assessment utilising an inverse problem approach,”
Artif. Intell. Med., vol. 54, no. 1, pp. 29–41, 2012.
[6] O. Troynikov and E. Ashayeri, “3D body scanning method for close-fitting garments in
sport and medical applications,” in Ergonomics Australia - HFESA 2011 Conference
Edition, 2011, pp. 11–16.
[7] E. Strait, G. McGimpsey, and T. C. T. Bradford, “Limb prosthetics services and devices,”
Bioeng. Inst. Cent. Neuroprosthetics Worcester Polytech. Inst., no. January, pp. 1–35,
2008.
[8] J. E. Sanders, S. G. Zachariah, A. K. Jacobsen, and J. R. Fergason, “Changes in interface
pressures and shear stresses over time on trans-tibial amputee subjects ambulating with
prosthetic limbs: Comparison of diurnal and six-month differences,” J. Biomech., vol. 38,
no. 8, pp. 1566–1573, 2005.
[9] J. Likitlersuang, M. J. Leineweber, and J. Andrysek, “Evaluating and improving the
performance of thin film force sensors within body and device interfaces,” Med. Eng.
Phys., vol. 48, pp. 206–211, 2017.
[10] Sensitronics LLC, “FSR 101 Force Sensing Resistor Theory and Applications,” 2017.
[11] J. R. Pearson, G. Holmgren, L. March, and K. Oberg, “Pressures in critical regions of the
below knee patellar tendon bearing prosthesis,” Bull. Prosthet. Res., no. 19, pp. 52–76,
1973.
[12] P. Laszczak et al., “Development and validation of a 3D-printed interfacial stress sensor
for prosthetic applications,” Med. Eng. Phys., vol. 37, no. 1, pp. 132–137, 2015.
[13] M. A. Razian and M. G. Pepper, “Design, development, and characteristics of an in-shoe
triaxial pressure measurement transducer utilizing a single element of piezoelectric
72
copolymer film,” IEEE Trans. Neural Syst. Rehabil. Eng., vol. 11, no. 3, pp. 288–293,
2003.
[14] R. K. Chen, Y. an Jin, J. Wensman, and A. Shih, “Additive manufacturing of custom
orthoses and prostheses-A review,” Addit. Manuf., vol. 12, pp. 77–89, 2016.
[15] P. Laszczak, L. Jiang, D. L. Bader, D. Moser, and S. Zahedi, “Development and validation
of a 3D-printed interfacial stress sensor for prosthetic applications,” Med. Eng. Phys., vol.
37, no. 1, pp. 132–137, 2015.
[16] B. Rogers, G. Bosker, M. Faustini, G. Walden, R. R. Neptune, and R. Crawford, “Case
report: variably compliant transtibial prosthetic socket fabricated using solid freeform
fabrication,” J. Prosthetics Orthot., vol. 20, no. 1, pp. 1–7, 2008.
[17] D. M. Sengeh and H. Herr, “A variable-impedance prosthetic socket for a transtibial
amputee designed from magnetic resonance imaging data,” J. Prosthetics Orthot., vol. 25,
no. 3, pp. 129–137, 2013.
[18] Y. Xu et al., “The Boom in 3D-Printed Sensor Technology.,” Sensors (Basel)., vol. 17, no.
5, 2017.
[19] J. T. Muth et al., “Embedded 3D printing of strain sensors within highly stretchable
elastomers,” Adv. Mater., vol. 26, no. 36, pp. 6307–6312, 2014.
[20] S.-Z. Guo, K. Qiu, F. Meng, S. H. Park, and M. C. McAlpine, “3D Printed Stretchable
Tactile Sensors,” Adv. Mater., vol. 29, no. 27, p. 1701218, 2017.
[21] S. B. Ge, X. C. Duan, P. Li, and T. Zheng, “A compliant translational joint based
force/displacement integrated sensor,” 2015, vol. 2015, no. CP676.
[22] V. Tiem, J. Groenesteijn, R. Sanders, and G. Krijnen, “3D Printed Bio-inspired Angular
Acceleration Sensor,” in Proc. IEEE Sensors, 2015, vol. 4, no. 5, pp. 1430–1433.
[23] Z. Shen, A. G. P. Kottapalli, V. Subramaniam, M. Asadnia, J. Miao, and M. Triantafyllou,
“Biomimetic flow sensors for biomedical flow sensing in intravenous tubes,” 2017.
[24] T. Beyrouthy, S. K. Al Kork, J. A. Korbane, and A. Abdulmonem, “EEG Mind controlled
Smart Prosthetic Arm,” 2016, pp. 404–409.
[25] C. Kousiatza and D. Karalekas, “In-situ monitoring of strain and temperature distributions
during fused deposition modeling process,” Mater. Des., vol. 97, pp. 400–406, 2016.
[26] L. Xiang, Z. Wang, Z. Liu, S. E. Weigum, Q. Yu, and M. Y. Chen, “Inkjet Printed
Flexible Biosensor Based on Graphene Field Effect Transistor,” IEEE Sens. J., vol. PP,
no. 99, 2016.
[27] N. Dakhil et al., “Is skin pressure a relevant factor for socket assessment in patients with
lower limb amputation?,” Technol. Heal. Care, pp. 1–9, Apr. 2019.
[28] G. Pirouzi, N. A. Abu Osman, A. Oshkour, S. Ali, H. Gholizadeh, and W. Wan Abas,
73
“Development of an Air Pneumatic Suspension System for Transtibial Prostheses,”
Sensors, vol. 14, no. 9, pp. 16754–16765, Sep. 2014.
[29] M. Monga, J. Premoli, N. Skemp, and W. Durfee, “Forearm compression by laparoscopic
hand-assist devices.,” J. Endourol., vol. 18, no. 7, pp. 654–6, Sep. 2004.
[30] S. D. Middleton, P. J. Jenkins, A. Y. Muir, R. E. Anakwe, and J. E. McEachan,
“Variability in local pressures under digital tourniquets,” J. Hand Surg. (European Vol.,
vol. 39, no. 6, pp. 637–641, Jul. 2014.
[31] S. Parmar, I. Khodasevych, and O. Troynikov, “Evaluation of Flexible Force Sensors for
Pressure Monitoring in Treatment of Chronic Venous Disorders,” Sensors, vol. 17, no. 8,
p. 1923, Aug. 2017.
[32] R. Ouckama and D. J. Pearsall, “Evaluation of a flexible force sensor for measurement of
helmet foam impact performance.,” J. Biomech., vol. 44, no. 5, pp. 904–9, Mar. 2011.
[33] R. S. Kearney, S. E. Lamb, J. Achten, N. R. Parsons, and M. L. Costa, “In-Shoe Plantar
Pressures Within Ankle-Foot Orthoses,” Am. J. Sports Med., vol. 39, no. 12, pp. 2679–
2685, 2011.
[34] A. Jule, B. M. Alistair Knott, and S. Mills, “Discriminative touch from pressure sensors,”
in 2015 6th International Conference on Automation, Robotics and Applications (ICARA),
2015, pp. 279–282.
[35] A. A. Polliack, R. C. Sieh, D. D. Craig, S. Landsberger, D. R. McNeil, and E. Ayyappa,
“Scientific validation of two commercial pressure sensor systems for prosthetic socket
fit,” Prosthet. Orthot. Int., vol. 24, no. 1, pp. 63–73, 2000.
[36] J. E. Sanders, C. H. Daly, and E. M. Burgess, “Clinical measurement of normal and shear
stresses on a trans-tibial stump: Characteristics of wave-form shapes during walking,”
Prosthet. Orthot. Int., vol. 17, no. 1, pp. 38–48, 1993.
[37] L. Frossard, J. Beck, M. Dillon, and E. John, “Development and Preliminary Testing of a
Device for the Direct Measurement of Forces and Moments in the Prosthetic Limb of
Transfemoral Amputees during Activities of Daily Living,” J. Prosthetics Orthot., vol. 15,
no. 4, pp. 135–142, 2003.
[38] N. A. Abu Osman, W. D. Spence, S. E. Solomonidis, J. P. Paul, and A. M. Weir, “The
patellar tendon bar! Is it a necessary feature?,” Med. Eng. Phys., vol. 32, no. 7, pp. 760–
765, Sep. 2010.
[39] I. Khodasevych, S. Parmar, and O. Troynikov, “Flexible Sensors for Pressure Therapy:
Effect of Substrate Curvature and Stiffness on Sensor Performance,” Sensors, vol. 17, no.
10, p. 2399, Oct. 2017.
[40] M. I. Tiwana, S. J. Redmond, and N. H. Lovell, “A review of tactile sensing technologies
with applications in biomedical engineering,” Sensors Actuators, A Phys., vol. 179, pp.
17–31, 2012.
74
[41] M. Ferguson-Pell, “Design criteria for the measurement of pressure at body/support
interface,” Eng. Med., vol. 9, no. 4, pp. 209–14, 1980.
[42] Peratech Ltd., “QTC ® SP200 Series Datasheet,” 2015.
[43] A. D. Lantada, P. Lafont, J. L. M. Sanz, J. M. Munoz-Guijosa, and J. E. Otero, “Quantum
tunnelling composites: Characterisation and modelling to promote their applications as
sensors,” Sensors Actuators, A Phys., vol. 164, no. 1–2, pp. 46–57, 2010.
[44] Tekscan, “Flexiforce ® Sensors Users Manual,” 2010.
[45] M. Ferguson-Pell, S. Hagisawa, and D. Bain, “Evaluation of a sensor for low interface
pressure applications,” Med. Eng. Phys., vol. 22, no. 9, pp. 657–663, 2001.
[46] T. R. Jensen, R. G. Radwint, and J. G. Webster, “A CONDUCTIVE POLYMER
SENSOR FOR MEASURING EXTERNAL FINGER FORCES,” 1991.
[47] J. Barton, S. Tedesco, T. Healy, and B. O’Flynn, “Potential for new smart knee device to
cut down on knee surgery recovery time,” Eng. J., 2017.
[48] T. Dumbleton et al., “Dynamic interface pressure distributions of two transtibial prosthetic
socket concepts,” J. Rehabil. Res. Dev., vol. 46, no. 3, p. 405, 2009.
[49] G. Pirouzi, N. A. Abu Osman, A. Eshraghi, S. Ali, H. Gholizadeh, and W. A. B. Wan
Abas, “Review of the socket design and interface pressure measurement for transtibial
prosthesis,” Sci. World J., vol. 2014, 2014.
[50] J. E. Sanders and C. H. Daly, “Normal and shear stresses on a residual limb in a prosthetic
socket during ambulation : Comparison of finite element results with experimental
measurements,” 1993.
[51] M. Zhang, A. . Turner-Smith, A. Tanner, and V. . Roberts, “Clinical investigation of the
pressure and shear stress on the trans-tibial stump with a prosthesis,” Med. Eng. Phys.,
vol. 20, no. 3, pp. 188–198, Apr. 1998.
[52] Ming Zhang and V. C. Roberts, “The effect of shear forces externally applied to skin
surface on underlying tissues.,” J. Biomed. Eng., vol. 15, no. 6, pp. 451–456, Nov. 1993.
[53] Nia Technologies Inc., “Technology,” 2018. [Online]. Available:
https://niatech.org/technology/.
[54] J. Zuniga et al., “Cyborg beast: a low-cost 3d-printed prosthetic hand for children with
upper-limb differences.,” BMC Res. Notes, vol. 8, p. 10, 2015.
[55] Y.-K. Lin, T.-S. Hsieh, L. Tsai, S.-H. Wang, and C.-C. Chiang, “Using three-dimensional
printing technology to produce a novel optical fiber Bragg grating pressure sensor,”
Sensors Mater., vol. 28, no. 5, pp. 389–394, 2016.
[56] M. Saari, B. Xia, B. Cox, P. S. Krueger, A. L. Cohen, and E. Richer, “Fabrication and
Analysis of a Composite 3D Printed Capacitive Force Sensor,” 3D Print. Addit. Manuf.,
75
vol. 3, no. 3, pp. 137–141, 2016.
[57] E. Suaste-Gómez, G. Rodríguez-Roldán, H. Reyes-Cruz, and O. Terán-Jiménez,
“Developing an ear prosthesis fabricated in polyvinylidene fluoride by a 3D printer with
sensory intrinsic properties of pressure and temperature,” Sensors (Switzerland), vol. 16,
no. 3, 2016.
[58] Wacker Chemie AG, “Interim Report January-June 2015,” München, Germany, 2015.
[59] J. H. Low, P. M. Khin, and C. H. Yeow, “A pressure-redistributing insole using soft
sensors and actuators,” in Proc. IEEE International Conference on Robotics and
Automation, 2015, pp. 2926–2930.
[60] E. R. Komi, J. R. Roberts, and S. J. Rothberg, “Evaluation of thin, flexible sensors for
time-resolved grip force measurement,” Proc. Inst. Mech. Eng. Part C J. Mech. Eng. Sci.,
vol. 221, no. 12, pp. 1687–1699, 2007.
[61] Interlink Electronics, “FSR Integration Guide and Evaluation Parts Catalog,” 2007.
[62] S. Derler, A. B. Spierings, and K. U. Schmitt, “Anatomical hip model for the mechanical
testing of hip protectors,” Med. Eng. Phys., vol. 27, no. 6, pp. 475–485, 2005.
[63] M. P. McGrath et al., “Development of a residuum/socket interface simulator for lower
limb prosthetics,” Proc. Inst. Mech. Eng. Part H J. Eng. Med., vol. 231, no. 3, pp. 235–
242, Mar. 2017.
[64] J. S. Schofield, K. R. Evans, J. S. Hebert, P. D. Marasco, and J. P. Carey, “The effect of
biomechanical variables on force sensitive resistor error: Implications for calibration and
improved accuracy,” J. Biomech., vol. 49, no. 5, pp. 786–792, Mar. 2016.
[65] A. Hollinger and M. M. Wanderley, “Evaluation of Commerical Force-Sensing
Resistors,” Montréal, Canada, 2006.
[66] Sensitronics LLC, “Half Inch ThruMode FSR.” [Online]. Available:
https://www.sensitronics.com/products-half-inch-thru-mode-fsr.php.
[67] J. E. Sanders, D. Lam, A. J. Dralle, and R. Okumura, “Interface pressures and shear
stresses at thirteen socket sites on two persons with transtibial amputation,” J. Rehabil.
Res. Dev., vol. 34, no. 1, pp. 19–43, Jan. 1997.
[68] F. Nabhani and J. Bamford, “Mechanical testing of hip protectors,” J. Mater. Process.
Technol., vol. 124, no. 3, pp. 311–318, Jun. 2002.
[69] N. J. Mills, “The Biomechanics of Hip Protectors,” Proc. Inst. Mech. Eng. Part H J. Eng.
Med., vol. 210, no. 4, pp. 259–266, Dec. 1996.
[70] A. Siddiqui, M. Braden, M. P. Patel, and S. Parker, “An experimental and theoretical
study of the effect of sample thickness on the Shore hardness of elastomers,” Dent.
Mater., vol. 26, no. 6, pp. 560–564, Jun. 2010.
76
[71] S. Blumentritt, “A new biomechanical method for determination of static prosthetic
alignment,” Prosthet. Orthot. Int., vol. 21, no. 2, pp. 107–113, 1997.
[72] M. E. Jones, J. R. Steel, G. M. Bashford, and I. R. Davidson, “Static versus dynamic
prosthetic weight bearing in elderly trans-tibial amputees,” Prosthet. Orthot. Int., vol. 21,
no. 2, pp. 100–106, 1997.
[73] P. Castellini, R. Montanini, and G. M. Revel, “Development of a film sensor for static and
dynamic force measurement,” Rev. Sci. Instrum., vol. 73, no. 9, p. 3378, 2002.
[74] L. Ren, R. K. Jones, and D. Howard, “Predictive modelling of human walking over a
complete gait cycle,” J. Biomech., vol. 40, no. 7, pp. 1567–1574, Jan. 2007.
[75] “QTC Single-Point Sensors,” Peratech Holdco Limited, 2017. [Online]. Available:
https://www.peratech.com/qtc-single-point-sensors/.
[76] S. Dhote, K. Behdinan, J. Andrysek, and J. Bian, “Experimental Investigation of Hybrid
Nickel-Carbon Black PDMS Conductive Composites,” In preparation, 2019.
[77] Orthotic & Prosthetic Lab Inc., “Prosthetic Instructions.” [Online]. Available:
https://www.oandplabinc.com/prosthetic-instructions.html.
[78] E. Al-Fakih, N. Arifin, G. Pirouzi, F. R. Mahamd Adikan, H. N. Shasmin, and N. A. Abu
Osman, “Optical fiber Bragg grating-instrumented silicone liner for interface pressure
measurement within prosthetic sockets of lower-limb amputees,” J. Biomed. Opt., vol. 22,
no. 08, p. 1, Aug. 2017.
[79] R. S. Hall, G. T. Desmoulin, and T. E. Milner, “A technique for conditioning and
calibrating force-sensing resistors for repeatable and reliable measurement of compressive
force,” J. Biomech., vol. 41, no. 16, pp. 3492–3495, Dec. 2008.
77
Appendices
Appendix A: Explaining the Deadband in Hysteresis Data
Figure A-1: Raw data and fitted curve for calibration data showing no collected data below 3.5 N.
Figure A-2: Raw resistance, calculated force, and applied force waveforms displaying deadband.
Applied Force
Calculated Force
Raw Resistance
78
Appendix B: QTC Prototype Sensor Inter-Sample Repeatability
To evaluate the inter-sample repeatability, a force-sweep from 0-10N was applied to two samples
of the QTC prototype sensor using the 15 mm diameter load tip attachment. No backing or puck
were used. Five trials were repeated per sample. Figure A-3 displays the force-resistance curves
for the five samples for each sample.
Figure A-3: Characteristic resistance vs. force curve for two QTC prototype sensor samples.
The coefficient of variation was calculated across the five trials of each sensor sample, as well as
across all 10 trials of both sensors to quantify the repeatability. While each sensor’s repeatability
was approximately 10% (9.3 and 10.2% for sensors 1 and 2, respectively), the inter-sample
repeatability was 20.5%. This value surpasses the 10% goal, indicating future work is required in
refining future iterations of the sensing prototype.
79
Appendix C: Additional Data
Table A-1: CV values calculated for each material, configuration and area for both Peratech and
Sensitronics sensors, using both matched-area and generalized-area calibration methods.
Material Config Area
Coefficient of Variation (%)
Peratech
Matched-Area
Peratech
Generalized-
Area
Sensitronics
Matched-Area
Sensitronics
Generalized-
Area
Extra Soft NBNP 05 1.3 15.1 11.5 14.2
08 0.8 0.8 7.7 7.7
15 5.6 38.1 13.7 53.5
25 3.9 55.0 8.9 76.3
NBYP 05 0.5 2.6 14.3 20.9
08 0.7 0.7 9.4 9.4
15 0.9 1.4 13.2 19.7
25 0.9 2.4 7.9 12.2
YBNP 05 1.3 14.9 10.3 16.6
08 0.7 0.7 11.3 11.3
15 6.0 49.6 22.2 69.7
25 2.4 44.3 6.8 50.4
YBYP 05 1.2 5.9 6.7 15.1
08 1.4 1.4 7.8 7.8
15 0.7 2.8 7.4 10.3
25 0.7 2.1 10.4 10.0
Soft 10A NBNP 05 0.6 12.2 7.1 13.2
08 0.8 0.8 6.7 6.7
15 17.5 42.6 9.2 81.5
25 16.9 46.7 6.9 66.8
NBYP 05 0.6 5.0 10.3 9.7
08 0.6 0.6 8.7 8.7
15 0.9 1.9 6.0 15.2
25 0.9 3.5 8.7 9.6
YBNP 05 0.9 16.7 12.6 15.7
08 1.0 1.0 9.5 9.5
15 8.1 47.7 8.4 72.7
25 13.6 59.3 14.1 45.6
YBYP 05 2.7 9.2 9.2 11.7
08 1.4 1.4 5.5 5.5
15 1.7 2.5 7.6 9.1
25 0.8 4.7 3.4 8.7
Soft 20A NBNP 05 0.6 12.0 4.9 16.4
08 0.9 0.9 3.6 3.6
15 21.2 39.0 8.8 31.8
25 6.1 49.9 26.9 22.4
NBYP 05 1.2 4.6 8.6 10.6
80
08 0.8 0.8 6.7 6.7
15 1.4 1.5 7.1 11.5
25 0.7 2.3 7.8 11.0
YBNP 05 0.9 14.3 6.4 14.4
08 0.9 0.9 4.7 4.7
15 12.1 63.9 6.2 25.6
25 5.6 43.4 29.3 21.1
YBYP 05 1.3 5.2 5.7 16.5
08 1.7 1.7 9.2 9.2
15 1.7 2.8 7.9 13.8
25 1.1 3.9 4.2 8.8