Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Design, dynamics, and control of miniature catalytic combustion engines and direct propane PEM fuel cells
(USC Thesis Other)
Design, dynamics, and control of miniature catalytic combustion engines and direct propane PEM fuel cells
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
Design, Dynamics, and Control of Miniature Catalytic-Combustion Engines
and Direct Propane PEM Fuel Cells
by
Fares Maimani
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(MECHANICAL ENGINEERING)
May 2024
Copyright 2024 Fares Maimani
Acknowledgments
First and foremost, I express my deepest gratitude to my advisor and defense committee
chair, Dr. Paul D. Ronney, for his exceptional mentorship both professionally and personally
since my first days at USC. As the chair of the Aerospace and Mechanical Engineering
department and later as my research advisor in the combustion physics lab, his guidance
have profoundly influenced my career. I will always follow the principle “Hit ’em where they
ain’t”.
I am thankful to Dr. Surya Prakash and his lab members, particularly Adam, Vicente,
JP, and Bo. Their generosity in sharing their time and expertise greatly enhanced my fuel
cells research.
My appreciation extends to Dr. Satwindar Sadhal for serving on my defense committee, and Drs. Mitul Luhar and Fokion Egolfopoulos for their roles in my qualifying exam
committee. Their insightful feedback significantly advanced my research.
I acknowledge my colleagues from the former AMSL lab at USC for their collaboration
on the microrobotics project, including Dr. Nestor Perez-Arancibia, Ariel, Xiufeng, and
Alberto. Ariel’s patience and thoughtfulness were invaluable as he guided through preparing
my first published paper.
Thanks are also due to my friends and collaborators in the combustion physics lab,
especially Dr. Eugene Kong, with whom I advanced the fuel cell project.
During my PhD, I had the privilege to engage in industry research projects. I am especially grateful to all my mentors at Meta for believing in me and providing opportunities
to broaden my research horizon. I am deeply indebted to Tim Otchy, who was consistently
supportive and instrumental in shaping me into a better research scientist.
The past few years in Los Angeles have been filled with joy, largely thanks to my incredible
friends who often also acted as informal collaborators. My friend Orazio, whom I met early
ii
in the PhD program, has inspired me to always strive for the best version of myself, whether
at work or at Gold’s Gym (Yeah Buddy!). My friend Alberto was always there willing to
help, whether by debugging code or making me the best pizza in town. My friends Javi,
Regina, Maria, Kaycee, Alex, Sara, and David have made the PhD journey more enjoyable
and made California feel like home with their love and unwavering support. I am proud and
grateful to have such dependable and inspiring individuals among my friends.
Finally, this journey would not have been possible without the unconditional support of
my family. My parents, Taghreed and Abed, my siblings, Ghaidaa, Yara, Jawad, and my
nephew Qusai, have been my steadfast support system. In their own ways, their constant
love and encouragement have made the distance from home more bearable. I am profoundly
thankful for their pride and belief in me.
iii
Table of Contents
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
Chapter 1:
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Chapter 2:
Miniature catalytic-combustion engine for millimeter-scale robotic actuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Design and fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Engine model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Experiment setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5 Characterization results and discussion . . . . . . . . . . . . . . . . . . . . . 21
2.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Chapter 3:
Dynamics of direct propane proton exchange membrane fuel cells . . . . 27
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.1.2 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.1.3 Summary of previous work in our lab . . . . . . . . . . . . . . . . . . 34
3.1.4 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2 Experiment methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2.1 Experiment apparatus . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2.2 Membrane electrode assembly (MEA) . . . . . . . . . . . . . . . . . . 42
3.2.3 Cell operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.3 OCV with research grade propane . . . . . . . . . . . . . . . . . . . . . . . . 46
3.4 Deactivation dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.4.1 Effect of current density . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.4.2 Effect of fuel flow rate . . . . . . . . . . . . . . . . . . . . . . . . . . 54
iv
3.4.3 Effect of catalyst loading . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.4.4 Effect of cell temperature . . . . . . . . . . . . . . . . . . . . . . . . 58
3.4.5 Deactivation dynamics at high cell temperatures . . . . . . . . . . . . 60
3.4.6 Summary of MEAs performance . . . . . . . . . . . . . . . . . . . . . 61
3.4.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.5 Activation dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.5.1 Effect of current density . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.5.2 Effect of cell voltage before reactivation . . . . . . . . . . . . . . . . . 65
3.5.3 Effect of time in full deactivation state . . . . . . . . . . . . . . . . . 67
3.5.4 Effect of total energy and total charge . . . . . . . . . . . . . . . . . 69
3.5.5 Effect of cell temperature . . . . . . . . . . . . . . . . . . . . . . . . 70
3.5.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.6 Effect of fuel additives on cell performance . . . . . . . . . . . . . . . . . . . 73
3.6.1 Effect of ethylene as a fuel additive . . . . . . . . . . . . . . . . . . . 73
3.6.2 Effect of hydrogen as a fuel additive . . . . . . . . . . . . . . . . . . . 80
Chapter 4:
Control of direct propane proton exchange membrane fuel cells . . . . . . 84
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.2 Dynamic model of the fuel cell . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.3 Formulating the control problem . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.4 On-off control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.5 Model predictive control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
Chapter 5:
Future Research Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
v
List of Tables
2.1 Biot Number for the radial and longitudinal dimensions in the system. . . . 12
2.2 Summary of simulation predictions for the miniature catalytic-combustion
engine at steady state operation for different frequencies. . . . . . . . . . . . 16
3.1 Comparison of the different MEAs used in the experiments. . . . . . . . . . . 42
4.1 Parameters for sigmoid-fit (Eq. 4.6) to rate constants for PEMFC dynamic
model (Eq. 4.5). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
vi
List of Figures
2.1 The miniature catalytic-combustion engine. (a) Close-up picture of the miniature engine. Shown in the dashed rectangle is the coated portion of the SMA
wire. (b) SEM image of the Pt coating layer on an SMA wire. The catalytic
coating layer is rough and porous and has a thickness of 18 µm. (c) Conceptual schematic that shows the allowed energy exchange paths in the thermal
model of the SMA-Pt system. . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Typical hysteresis transformation curve for an SMA wire under constant stress
during heating and cooling. As: austenite start, Af : austenite finish, Ms:
martensite start, Mf : martensite finish. . . . . . . . . . . . . . . . . . . . . . 5
2.3 Conceptual design of a robot utilizing the miniature catalytic-combustion engine for actuation. A Fuel tank releases fuel periodically to react with the
catalytic layer coating the SMA wire. The actuator body provides preloads
the SMA wire. The heat of reaction and the preload stress causes the SMA
wire to contract and recover its shape cyclically. Frictional anisotropic legs
converts the cyclic strain deformation of the SMA to forward motion of the
robot. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.4 Design of the miniature catalytic-combustion engine. (a) A segment of SMA
wires are coated with Pt. When no fuel is present to initiate catalytic combustion, the actuator stays in its relaxed position. (b) When a gas source
introduces fuel gas around the Pt coating layer, catalytic combustion starts.
The heat released from the reaction results in the shortening of the SMA wire,
which creates bending motion on the actuator. (c) The gas source supplies an
air flow to increase the cooling rate by convection. As a result, the actuator
goes back to its relaxed position. . . . . . . . . . . . . . . . . . . . . . . . . 9
vii
2.5 Fabrication of the SMA based actuators Step 1: A jig of FR4 is pre-cut using
a high resolution laser cutter. Step 2: An SMA wire is looped in the orifices
of the jig and secured at the ends using glue and knots. The SMA wire must
be kept in tension. Step 3: Using the tabs in the jig, a carbon fiber layer is
glued to the back of the jig. Step 4: The jig is aligned and cut using the high
resolution laser cutter. Step 5: A thermally conductive paste is added to the
surface of the SMA. A brush is used to manually add the paste evenly in a
section of the wire. Step 6: Pt powder is poured over the actuator. The Pt
powder coats the SMA surface area with where the paste was applied. Step
7: The actuator is ready to be used. . . . . . . . . . . . . . . . . . . . . . . . 10
2.6 Plots of the Platinum coating Layer (Pt) and the SMA wire temperatures
obtained from the numerical model for a specified H2 pulse train. In all simulations, the frequency of the fuel pulse train was specified and the duty cycle
was fixed at 25%. The amplitude of the fuel pulse train was chosen to limit
the maximum SMA temperature to 90◦C. As the frequency increases, there
is less time available for cooling, and as a result, the steady state temperature
range decreases. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.7 Simulation results of the tip displacement for the miniature catalytic-combustion
engine. The displacement range decreases as the frequency of operation increases. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.8 Experimental Characterization Setup. (a.1) Gas flow sensors. (a.2) Solenoid
proportional valves connected to the sensors upstream and to the mixing junction downstream. The valves are placed on a 3D printed platform. (a.3) Laser
displacement sensor. (b) A close-up of the design of the 3D printed mixing
junction and actuator base. The mixing junction is shown in blue. The gas
flows are mixed in the junction and redirected to a nozzle that faces the actuator shown in orange. The actuator is held in place by a small piece of tape
(not shown) on one end, while the other end of the actuator is free to move and
perform work. (c) The closed loop feedback control scheme for the fuel delivery system is shown inside the dashed red box. The overall characterization
setup utilizes the displayed open loop feedforward control scheme. . . . . . . 18
2.9 Plots showing the linear displacement of the actuator tip during steady state
operation for a given fuel flow rate. The actuator is displacing a 2.15 g weight
against gravity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
viii
2.10 Frames from a video showing a full operational cycle of the catalytic actuator.
The time between frames (a) and (e) is 1 s. A red vertical line was added in
the same position in all frames to make the displacement of the actuator tip
clearer. The red dot on the actuator tip is the reflection of the displacement
laser sensor. (a) The actuator is hot and fully contracted. This is the beginning of the cooling phase. (b) As the actuator cools down by convective
and conductive forces it starts to expand, returning the actuator to its resting
position. At this point, the fuel gas reaches the catalytic layer and starts
reacting. This can be deducted from the slightly red glow appearing on the
catalyst’s surface. (c) The actuator fully reaches the relaxed position and
the catalytic combustion reaction is at its maximum rate. This is evident by
the red glow seen on the catalytic layer. (d) As the heat transfers from the
catalytic layer to the SMA wire and increases its temperature, the actuator
starts to bend. Some fuel gas is still reacting on the catalyst surface. (e) The
SMA temperature is at its maximum and the actuator is fully bent and is at
the beginning of the actuation cycle again. . . . . . . . . . . . . . . . . . . . 24
2.11 Frames from a video showing the actuator lifting a 4.55 g mass 0.91 mm at a
frequency of 1 Hz. A red line is added to signify the movement of the weight.
(a) The actuator is in its resting position. (b) The actuator is fully bent
achieving its maximum displacement and 39.5 µJ of work. . . . . . . . . . . . 25
3.1 Typical operation mode of a PEM fuel cell using hydrogen fuel and air as
reactants and generating electricity, heat, and water as products. . . . . . . . 29
3.2 Effect of ethylene impurity concentration on power production at constant
current of 36 mA/cm2
(for 0 PPM, the cell was first “ignited” with ethylene
then the ethylene flow was stopped). Note the logarithmic time scale [58]. . . 36
3.3 Constant current (galvanostatic) and “On-Off” modes of a 25 cm2 direct propane
PEMFC. The flow rate was 1.2 L/min and 0.8 L/min for propane and oxygen,
respectively. The cell temperature was held at 80 ◦C. For the On-Off mode,
the current was applied for 20 seconds and then shut off for 5 seconds, repeatedly [58]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4 Conceptual diagram of the fuel cell testing apparatus we built. . . . . . . . . 39
3.5 Picture of the main components of the fuel cell testing apparatus we built in
the lab. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.6 Schematic drawing of a single fuel cell assembly apparatus [58]. . . . . . . . 42
3.7 Screenshot of the LabVIEW UI used to perform fuel cell testing. . . . . . . . 43
ix
3.8 The carbon paper with catalyst painted on it after it dried and before it
is pressed onto the Nafion membrane to form MEA PtB CPL . The carbon
paper on the left had a loading of 12.2 mg/cm2 and the carbon paper on the
right had a loading of 9.2 mg/cm2
. . . . . . . . . . . . . . . . . . . . . . . . . 45
3.9 MEA PtC FCS OCV dynamics for different propane flow rates with 0.5 mg/cm2
catalyst loading. The fuel flow was started at time 0 and the voltage was measured until it stabilized. The OCV was observed to be stable for more than
3000 s. MEA PtC FCS with 1 mg/cm2 and 1.5 mg/cm2
catalyst loadings exhibited similar OCV dynamics and stability. . . . . . . . . . . . . . . . . . . 47
3.10 MEA PtB FCS OCV dynamics with 4 mg/cm2
catalyst loading. The fuel
flow was started at time 0 and the voltage was measured until it stabilized.
The observed voltage oscillations were confirmed to correspond to the fuel
cell temperature oscillations, which were caused the on-off cell temperature
controller’s non-zero dead-band. . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.11 MEA PtB CPL open circuit voltage (OCV) dynamics with 12.2 mg/cm2 and
9.2 mg/cm2
catalyst loading on the anode and cathode sides, respectively.The
fuel flow was started at time 0 and the voltage was measured until it stabilized. 48
3.12 MEA PtB CPL open circuit voltage (OCV) as a function of cell temperature. 49
3.13 Effect of the current density on the deactivation dynamics for MEA PtB CPL
with 12.2 mg/cm2 and 9.2 mg/cm2
loadings on the anode and cathode sides,
respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.14 Effect of the current density on the deactivation dynamics for MEA PtC FCS
with 1.5 mg/cm2
loading. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.15 Effect of the current density on the deactivation dynamics for MEA PtB FCS
with 4 mg/cm2
loading. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.16 The effect of propane fuel flow rate at the anode on the deactivation dynamics
of a 0.5 mg/cm2 MEA PtC FCS under a constant load of 3 mA/cm2
. (a)
Anode flow consisting solely of propane, varied as indicated in the legend, with
a constant cathode flow rate of 0.8 SLM oxygen. (b) Anode flow consisting
solely of propane, adjusted as per the legend, with oxygen cathode flow rate
also varied to maintain an anode:cathode flow rate ratio of 1.5. (c) Anode
flow composed of propane and nitrogen; propane flow rate varied as shown in
the legend, with nitrogen supplementing the anode flow to maintain a total
flow rate of 1.2 SLM, keeping the cathode flow rate constant at 0.8 SLM oxygen. 55
3.17 The effect of catalyst loading on the deactivation dynamics for MEA PtC FCS
for two different loading’s: (a) 0.5 mg/cm2 and (b) 1.5 mg/cm2
. . . . . . . . . 58
x
3.18 The effect of cell temperature on the deactivation dynamics for 0.5 mg/cm2
MEA PtC FCS and 12.2 mg/cm2 MEA PtB CPL under a constant current
load of 3 mA/cm2 and 30 mA/cm2
for (a) and (b), respectively. Note the
logarithmic timescale. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.19 Effect of the current density on the deactivation dynamics for MEA PtB CPL
with 12.2 mg/cm2 anode catalyst loading and operating at 104 ◦C. Note the
logarithmic timescale. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.20 Effect of the current density during deactivation on the activation dynamics
for MEA PtB CPL with 12.2 mg/cm2 anode catalyst loading and operating
at 85 ◦C. Note the logarithmic timescale. . . . . . . . . . . . . . . . . . . . . 64
3.21 Effect of the current density during deactivation on the activation rate and activation duration for MEA PtB CPL with 12.2 mg/cm2 anode catalyst loading
and operating at 85 ◦C. Both the activation rate and the activation duration
were calculated between the lowest observed voltage value and 0.6 V. . . . . 65
3.22 Effect of the cell’s voltage before activation is initiated on the activation dynamics for MEA PtB CPL with 12.2 mg/cm2 anode catalyst loading and operating at 85 ◦C. Note the logarithmic timescale. . . . . . . . . . . . . . . . . 66
3.23 Effect of the cell’s voltage before activation is initiated on the activation rate
and activation duration for MEA PtB CPL with 12.2 mg/cm2 anode catalyst
loading and operating at 85 ◦C. Both the activation rate and the activation
duration were calculated between the voltage value specified on the x-axis and
0.6 V. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.24 Effect of the duration the cell spends in deactivated state on the activation
dynamics for MEA PtB CPL with 12.2 mg/cm2 anode catalyst loading and
operating at 85 ◦C. Note the logarithmic timescale. . . . . . . . . . . . . . . 68
3.25 Effect of the duration the cell spends in deactivated state on on the activation rate and activation duration for MEA PtB CPL with 12.2 mg/cm2 anode
catalyst loading and operating at 85 ◦C. Both the activation rate and the activation duration were calculated between the lowest voltage value observed
and and 0.6 V. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.26 Effect of the total power produced by the cell on the activation dynamics
for MEA PtB CPL with 12.2 mg/cm2 anode catalyst loading and operating
at 85 ◦C. In (a), the cell was operated under a constant current load of
10 mA/cm2
. Note the logarithmic time scale in (b). . . . . . . . . . . . . . . 70
xi
3.27 Effect of the total power and total charge produced by the cell on the activation duration for MEA PtB CPL with 12.2 mg/cm2 anode catalyst loading
and operating at 85 ◦C. The activation duration were calculated between the
lowest voltage value observed and 0.6 V. . . . . . . . . . . . . . . . . . . . . 71
3.28 Effect the cell temperature on the activation dynamics for MEA PtB CPL
with 12.2 mg/cm2 anode catalyst loading. The cell was deactivated using a
constant current load of 30 mA/cm2
. Note the logarithmic timescale. . . . . 72
3.29 Effect of the cell temperature on the activation rate and activation duration for
MEA PtB CPL with 12.2 mg/cm2 anode catalyst loading. Both the activation
rate and the activation duration were calculated between the lowest voltage
value observed and 0.6 V. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.30 Effect of adding 3 SCCM of ethylene (2500 PPM) to the anode fuel stream of
the MEA PtB CPL on the cell’s OCV. The anode and cathode flow streams
were 1.2 SLM propane and 0.8 SLM oxygen, respectively. (a) The ethylene
flow rate over time; we removed the ethylene flow rate temporarily around
t = 300 s. (b) The cell’s OCV. . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.31 Ethylene in Propane. The effect of adding different quantities of ethylene, as indicated in the sub-figures’ labels, to the anode flow rate of 1.2 SLM
propane on the deactivation dynamics for MEA PtB CPL under constant current load. The current loads are specified in the sub-figures’ legends. The
cathode flow rate was 0.8 SLM oxygen. . . . . . . . . . . . . . . . . . . . . . 77
3.32 Ethylene in Nitrogen. he effect of adding different quantities of ethylene,
as indicated in the sub-figures’ labels, to an inert anode flow rate of 1.2 SLM
nitrogen on the deactivation dynamics for MEA PtB CPL under constant
current load. The current loads are specified in the sub-figures’ legends. The
cathode flow rate was 0.8 SLM oxygen. . . . . . . . . . . . . . . . . . . . . . 77
3.33 Polarization curve for MEA PtB CPL operating on an anode flow rate of
1.2 SLM propane and 18 SCCM ethylene, and a cathode flow rate of 0.8 SLM
oxygen. The polarization curve is parameterized in time, to demonstrate the
partial deactivation dynamics at each current density value. . . . . . . . . . 78
xii
3.34 Comparison of our results with previous work on the impact of ethylene additives on direct propane PEMFC. (a) Results reproduced from [59] demonstrating the influence of ethylene concentration on power generation at 36 mA/cm2
(for 0 PPM, the cell was initially activated with a 5 s exposure to 2000 PPM of
ethylene in the fuel stream). They employed a direct propane PEMFC similar
to ours, operating at 80 ◦C with an anode flow rate of 1.2 SLM research-grade
propane. (b) Our results from the MEA PtB CPL illustrating the impact
of ethylene concentration on deactivation dynamics for a cell operating at
36 mA/cm2 with an anode flow rate of 1.2 SLM nitrogen. The nitrogen is
inert and does not contribute to power production. . . . . . . . . . . . . . . 79
3.35 Effect of ethylene dilution on the deactivation dynamics of MEA PtB CPL
at 85 ◦C. The deactivation was assessed under two scenarios: non-diluted
ethylene with an anode flow rate of 1.2 SLM of pure ethylene, and diluted
ethylene, consisting of a 1.5% ethylene in nitrogen mixture. A constant current
load of 36 mA/cm2 was maintained during deactivation. Note the logarithmic
time-scale. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.36 The effect of adding 18 SCCM of hydrogen to an anode flow rate of 1.2 SLM of
(a) propane or (b) nitrogen on the power dynamics for MEA PtC FCS with
1.5 mg/cm2 PtC loading under constant current loads. The current loads are
specified in the legends. The cathode flow rate was 0.8 SLM oxygen. . . . . . 81
3.37 Polarization curve for the MEA PtC FCS with 1.5 mg/cm2 PtC loading operating with an anode flow mixture of hydrogen-nitrogen or hydrogen-propane
with different mixing ratios. The mixing ratios are specified in the legend.
The cathode flow rate was 0.8 SLM oxygen. . . . . . . . . . . . . . . . . . . . 83
3.38 Polarization curve for the MEA PtB FCS with 4 mg/cm2 PtB loading operating with an anode flow mixture of hydrogen-nitrogen or hydrogen-propane
with different mixing ratios. The mixing ratios are specified in the legend.
The cathode flow rate was 0.8 SLM oxygen. . . . . . . . . . . . . . . . . . . . 83
4.1 Visualization of a symmetric sigmoid function, sigmoid(x) = 1
1+exp (x)
. . . . . 89
4.2 A comparison of the dynamic propane PEMFC model and the experimental
results. For both results, a current load of I = 30 mA/cm2 was applied on
the cell until full deactivation was reached(V < 0.1 or Xa < 0.1), then the
current load was removed, by setting I = 0 mA/cm2
. (a) Shows the model
prediction for Xa. We set Xa,0 = 0.8 to better match experimental results.
(b) Shows the experimental voltage dynamics as the cell goes through one
deactivation-activation cycle. . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.3 General feedback control architecture for the direct propane PEMFC. . . . . 91
xiii
4.4 The design of an On-off current controller with hysteresis. . . . . . . . . . . 93
4.5 Simulation of an on-off controller performance using the architecture from
figure 4.4. We set Ih = 30 mA/cm2
, Il = 0 mA/cm2
, Vl = Va = 0.3 V, and
Vh = Vb = 0.6 V. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.6 Experimental results for implementing the on-off controller. We set Ih =
30 mA/cm2
, Il = 0 mA/cm2
, Vl = 0.1 V, and Vh = 0.7 V. (a) Shows the
voltage dynamics, (b) shows the power dynamics, and (c) Shows the the period
for one on-off cycle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.7 Experimental results for implementing the on-off controller. For all experiments, we set Vl = 0.1 V, and Vh = 0.7 V. For (a, b), Ih = 40 mA/cm2
,
Il = 0 mA/cm2
. For (c, d) Ih = 25 mA/cm2
, Il = 0 mA/cm2
. For (e, f),
Ih = 30 mA/cm2
, Il = 2 mA/cm2
. . . . . . . . . . . . . . . . . . . . . . . . . 96
4.8 Experimental results for implementing the on-off controller on a high temperature cell operating at 101 ◦C. We set Ih = 55 mA/cm2
, Il = 0 mA/cm2
,
Vl = 0.1 V, and Vh = 0.7 V. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.9 Simplified block diagram of a MPC-based control loop [77]. . . . . . . . . . . 98
4.10 Simulation of the MPC controller performance using formulation in Eq. 4.15. 101
4.11 Simulation of the MPC Controller robustness in a PEMFC dynamical system
with noise-perturbed rate constants (Ki). . . . . . . . . . . . . . . . . . . . . 102
4.12 Experimental implementation of the MPC controller with short time horizons,
as specified in the captions of the subfigures. For both controllers, the solution’s time step was maintained at 0.25 s, corresponding to a sampling rate of
4 Hz. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
xiv
Abstract
This work investigates the use of chemical fuels like hydrogen and propane for small scale
power generation. Chemical fuels’ high specific energy makes them promising for applications with stringent power system requirements. We focus on two projects: one employing
hydrogen with catalytic combustion for powering microrobots, and another utilizing propane
in proton exchange membrane fuel cells (PEMFC) for portable power.
Microrobots at the subcentimeter scale have the potential to perform useful complex
tasks if they were to become energy independent and could operate autonomously. The vast
majority of current microrobotic systems lack the ability to carry sufficient onboard power
to operate and, therefore, remain tethered to stationary sources of energy in laboratory
environments. Recent published work demonstrated that chemical fuels can react under
feedback control on the surfaces of tensioned shape-memory alloy (SMA) nickel-titanium
(NiTi) wires coated with platinum (Pt) catalyst. Combining catalytic combustion of fuels
with high energy densities with the high work densities of SMA wires is a promising approach
to provide onboard power to microrobots. In this work, we present a novel 7-mg SMAbased miniature catalytic-combustion engine for millimeter-scale robotic actuation that is
composed of a looped NiTi-Pt composite wire with a core diameter of 38 µm and a flat
carbon-fiber beam with a length of 13 mm. This beam acts as a leaf spring during operation.
The proposed design of the engine has a flat and narrow geometry, functions according to a
periodic-unimorph actuation mode, and can operate at frequencies as high as 6Hz and lift
650 times its own weight while functioning at 1Hz, thus producing 39.5 µW of average power
in the process. For the purposes of design and analysis, we derived a model of the heat
transfer processes involved during actuation, which combined with a Preisach-model-based
description of the SMA wire dynamics, enabled us to numerically simulate the response of
the miniature system, and thus predict its performance in terms of frequency and actuation
xv
output. The suitability for microrobotics and functionality of the proposed approach is
demonstrated through experimental results using a custom-built fast-response high-precision
system of fuel delivery.
Low-temperature direct hydrocarbon PEMFC offer a viable solution for portable power
generation, with capacities under 100W. These cells provide a practical alternative to hydrogen PEMFC due to the ease of storage and handling of hydrocarbons. Moreover, they exhibit
a potential energy density advantage of 10 to 50 times over traditional batteries. Despite
their potential, challenges such as low power densities and the occurrence of an extinction
phenomenon—wherein the cell stops producing power—remain. To address these issues, we
developed a custom experimental setup to explore the dynamics and control of direct propane
PEMFC, utilizing three distinct membrane electrode assemblies with varying platinum catalyst configurations. Our findings diverge from prior studies, showing sustained open circuit
voltage and power generation capabilities with high purity propane alone. The study investigates the deactivation and activation dynamics under varied conditions, including current
density, fuel flow rate, catalyst specifics, cell temperature, and operational modes. Current
density was found to be the critical factor influencing cell behavior, while increasing the
temperature significantly enhanced performance by accelerating activation and decelerating
deactivation. Experimentally, we demonstrated that the direct propane PEMFC achieves
power densities on the scale of 10 mW/cm2 at 85 ◦C. Additionally, we examined the impact
of ethylene and hydrogen additives in propane on cell dynamics. Based on the experiment
observations, we formulated a dynamical model to capture the direct propane cell dynamics. This model informed the development and evaluation of two control strategies: on-off
control and model predictive control, aimed at enhancing average power output. Simulation
and experimental validation of these strategies demonstrated an increase in average power
density by 59%, underscoring the effectiveness of active feedback control in optimizing direct
propane PEMFC performance.
xvi
Chapter 1
Introduction
The specific energy and specific power are critical factors to consider when evaluating a
power generation system. In this context, the specific energy is defined as the total energy
the system can provide divided by its total mass, while the specific power is defined as the
maximum instantaneous or average power the system can provide divided by its total mass.
The total mass of the system includes the fuel storage, the energy conversion device, and
any auxiliary components.
In certain portable power generation applications, the specific energy and specific power
requirements are strict and the standard battery technologies of today cannot efficiently meet
those requirements. In this work, I examine two cases of such applications: autonomous microrobots and portable electronics. In both applications we propose and investigate methods
of utilising chemical fuels (such as hydrogen or propane) to meet the design requirements.
The main motivation behind using chemical fuels is to exploit their significantly higher specific energy and power, compared to batteries. In this proposal, we present two projects that
aim to provide innovative small scale power generation solutions using chemical fuels.
In Chapter 2, we present a miniature catalytic-combustion engine for millimeter-scale
robotic actuation. For autonomous microrobots in the sub-centimeter scale, having onboard power generation is essential for proper functioning. However, supplying and storing
the energy on-board is still a challenging problem and is an active area of research. Standard
batteries cannot be used to meet the energy and power requirements of the microrobot
1
because of their relatively low specific energy and power. We demonstrate a path toward
using hydrogen fuel as an energy source to provide actuation power to crawling microrobots.
The hydrogen chemical energy is converted to mechanical actuation energy through the use
of a miniature engine that utilizes shape memory alloy wires and catalytic combustion. We
discuss the modelling, design, fabrication, and characterization results of the engine.
In Chapter 3, we present our study on direct propane proton exchange membrane fuel
cells (PEMFC). While hydrogen PEMFC are widely studied and used, there has been less
attention on direct propane PEMFC. Because propane is easier to store and handle than
hydrogen, direct propane PEMFC can be an attractive portable power source for applications such as portable electronics in the scale of 100W or less. Previous work in our lab
investigated propane PEMFC and observed an extinction phenomena that can be prevented
by using unsaturated hydrocarbon additives in the fuel stream. In this chapter, we discuss
the custom-built fuel cell testing apparatus and present our investigation on the deactivation and activation dynamics of a PEMFC running on high purity propane. Additionally,
we present results on the effect of different fuel mixtures on the cell’s performance.
In Chapter 4, we develop a dynamic model for the direct propane PEMFC. This model
supports the design and evaluation of two control strategies: on-off control, and model
predictive control. We assess these strategies through simulations based on the dynamic
model and through experiments conducted with our fuel cell testing apparatus.
In Chapter 5, we focus on two key areas for future research: investigating PEMFC deactivation mechanisms and integrating our power sources into real-world applications. Understanding why direct propane PEMFC deactivate is crucial for improving their applicability
and performance. Additionally, we discuss the need to develop auxiliary systems such as fuel
delivery and thermal management systems before deploying the small-scale power generation
technologies discussed in real-world applications.
2
Chapter 2
Miniature catalytic-combustion engine
for millimeter-scale robotic actuation
2.1 Introduction
Fully autonomous microrobots at the subcentimeter scale will revolutionize many fields such
as surveillance, drug-delivery, search and rescue, and artificial pollination. However, due to a
lack of adequate power systems that can operate the robots in a tetherless and autonomous
way, most microrobots are usually only operable inside labs where they are connected to stationary power systems. Fully autonomous microrobots must be able to operate and achieve
a goal without depending on any external system [1]. We define untethered microrobots
as robots that can operate without being connected through cables to any external system.
Untethered microrobots may connect wirelessly to external systems for power or control. In
this article, we use the term microrobot to describe miniaturized robots with characteristic
lengths at the subcentimeter scale. To achieve untethered operation, microrobots need power
systems that can be added on board, or external power systems that can deliver power wirelessly. In the case of onboard power, the system needs to have adequate specific energy and
specific power ratings to supply the microrobot with the energy required for its operation. In
this article, we demonstrate combining catalytic combustion of fuels with high energy den3
Figure 2.1: The miniature catalytic-combustion engine. (a) Close-up picture of the miniature
engine. Shown in the dashed rectangle is the coated portion of the SMA wire. (b) SEM
image of the Pt coating layer on an SMA wire. The catalytic coating layer is rough and
porous and has a thickness of 18 µm. (c) Conceptual schematic that shows the allowed
energy exchange paths in the thermal model of the SMA-Pt system.
sities with the high work densities of shape-memory allow wires as an approach to provide
onboard power to microrobots (Fig. 2.1). The traditional technologies and approaches used
in the literature cannot easily meet the requirements for powering autonomous microrobots.
Wireless external power sources have been investigated to different levels of success to
build untethered microrobots. Different research groups have used lasers [2], solar power
[3], vibration fields [4], and magnetic fields [5, 6] to power and control microrobots. While
promising, wireless power sources do not allow for full autonomy, and, as a result, they limit
the scope of the environment in which microrobots can operate in while also limiting their
the versatility and robustness. Furthermore, powering and controlling multiple microrobots
independently through wireless power fields is challenging. To design a robust and versatile
microrobot, we believe that a power source with high specific energy and specific power must
be carried on board the microrobot.
The majority of microrobots with onboard power architectures proposed in the literature
utilizes actuation systems that converts electrical energy to mechanical energy in order to
locomote. The main technologies used in those designs are piezoelectric effect [7, 8], electrostatic actuators [9], and shape-memory alloys (SMAs) [10, 11]. Piezoelectric actuation
systems are robust and operate at high frequencies, but they require high voltage electronics
4
Figure 2.2: Typical hysteresis transformation curve for an SMA wire under constant stress
during heating and cooling. As: austenite start, Af : austenite finish, Ms: martensite start,
Mf : martensite finish.
Figure 2.3: Conceptual design of a robot utilizing the miniature catalytic-combustion engine
for actuation. A Fuel tank releases fuel periodically to react with the catalytic layer coating
the SMA wire. The actuator body provides preloads the SMA wire. The heat of reaction and
the preload stress causes the SMA wire to contract and recover its shape cyclically. Frictional
anisotropic legs converts the cyclic strain deformation of the SMA to forward motion of the
robot.
[12]. Electrostatic actuators are energy efficient and can operate at high frequencies, but
they also suffer from low energy densities and need high voltages [12]. SMAs are smart materials, usually made of nickel-titanium (NiTi), that have crystal structures and can strain
in response to temperature and stress changes. Tensioned SMA wires contract when heated
and recover their original shape when cooled down while exhibiting a shape-memory effect
[13] (Fig. 2.2). SMA-based actuators have high power densities and require relatively low
voltages, but their operation bandwidth is limited by the time required for cooling the SMA
back to its original shape [14].
One approach to provide electrical input to microrobotic actuation systems is to use
onboard batteries. State of the art battery technology cannot easily power microrobots
5
because of their limited specific energy and specific power ratings. Batteries have been
successfully used in autonomous robots with length scales as small as 4 cm [15]. However,
for smaller robots at the subcentimeter scale, using batteries is difficult because of the relative
large battery mass and the required components to interface between the battery and the
actuation mechanism. For example, the SMALLBug shown in Ref. [16] is a 1.2 cm, 30 mg
crawling microrobot that uses an SMA actuator and requires 135 mW of average power to
operate at a frequency and speed of 20 Hz and 17 mm s−1
, respectively. To operate the
SMALLBug for 10 minutes using state of the art Li-ion batteries with a specific energy and
specific power of 0.875 MJ kg−1
and 340W kg−1
, respectively, the required battery will weigh
400 mg, or 13 times the robot weight [17]. Additional weight is further required to meet the
instantaneous power requirements of the SMALLBug, to interface between the battery and
the actuation system, and to power any additional onboard systems. This calculation also
assumes that batteries’ specific energy and specific power do not decrease at small scales.
Beside batteries, other novel approaches such as using living cells on board to power and
actuate microrobots have also been investigated [18].
A more promising approach toward having onboard power in microrobots is to use chemical fuels, such as hydrogen or hydrocarbons, as an alternative to batteries. This approach
is more promising because chemical fuels have high specific energy (45 MJ kg−1
for Butane)
and high specific power when combined with an appropriate heat engine. Miniaturizing a
traditional heat engine is difficult because of fabrication challenges and efficiency reductions
due to friction and heat losses. One concept that has been investigated is to design miniature
heat engines using SMAs. The shape-memory effect of SMA wires allows them to be used in
an analogous way to working fluids in heat engines to convert chemical energy to mechanical
work.
Early work investigated using SMA coils to build large heat engines [19], but most efforts
failed due to scaling issues. More recently, SMA engines and actuators have been built
successfully at smaller scales in many fields such as robotics [20, 21], MEMS [22], wireless
6
sensors [23], and microfluidics [24].
The majority of SMA actuators use electricity to heat up the SMA through Joule heating
and use ambient air for cooling. The majority of SMA heat engines use a stationary hot
source, such as hot water, to heat up the SMA. Those designs do not address how the hot
source can be powered at the relevant scale. A promising approach to heat up the SMA has
been to use Platinum (Pt) catalyst to perform catalytic combustion of chemical fuels on or
near the SMA. Early attempts demonstrated the concept but only succeeded at achieving
slow actuation frequencies (<0.1 Hz) [25, 26]. Other efforts investigated catalytically burning
liquid propellant fuels to actuate microrobots by utilizing the hot product gases of combustion
[27]. Recent work successfully demonstrated that SMA wires coated with Pt can dependably
actuate microrobots using hydrogen, methanol, or butane with frequencies up to 1 Hz [20,
28]. The USC Autonomous Microrobotic Systems Lab (AMSL) successfully demonstrated
the RoBeetle, an autonomous 1.5 cm microrobotic crawler that catalytically burns methanol
on a Pt-coated SMA wire to move with speeds up to 0.76 mm s−1
[20].
In this work, we continue the previous efforts done in AMSL to develop high frequency microrobotic SMA actuation systems. Inspired by the fast actuation technology of the SMALLBug and the catalytic combustion innovations of the RoBeetle, we propose a novel 7 mg
miniature catalytic-combustion engine for millimeter-scale robotic actuation and demonstrate its operation with frequencies up to 6 Hz (Fig. 2.1a). For the rest of the chapter,
we use the words engine and actuator interchangeably to describe the proposed SMA-based
device, which converts chemical energy to mechanical work.
To design our miniature catalytic-combustion engine, we started from the actuator presented in Ref. [16] and modified it to be powered by hydrogen fuel. We performed thermal
and dynamic analysis on the miniature engine to demonstrate the design validity and model
the engine’s performance. To test the engine and its control strategy, we built an experimental characterization setup and an advanced fuel delivery system that can be used to characterize engines and design control strategies. The presented miniature engine demonstrates
7
a path toward improving the design and increasing the frequency of SMA-based miniature
catalytic-combustion engines that can potentially be used in autonomous microrobots.
The conceptual design illustrated in Fig. 2.3 demonstrates how the proposed miniature
catalytic-combustion engine can be implemented in a microrobotic application. The proposed
actuator and a fuel tank are attached to a robotic structure similar to the one presented in
[16]. The tank is preprogrammed to releases fuel periodically. The fuel reacts on the catalytic
surface causing the SMA wire temperature to oscillate. The actuator body acts as a spring
to preload the SMA wire. The preload stress and the temperature oscillation causes the
SMA wire to contract and expand repeatedly. The frictionally anisotropic legs of the robot
convert the cyclic strain deformations of the SMA wire to a forward motion of the robot. The
characterization setup we built can be used to test control and fuel delivery strategies. Then,
the preprogrammed, onboard tank and valve system can deliver the fuel to the actuator in a
way that achieves the desired functional and performative requirements of the microrobot.
The rest of the chapter is organized as follows. First, we discuss the fabrication and
design process of the novel miniature catalytic-combustion engines. Then, we present the
thermal and dynamic analysis of the proposed miniature engine. After, we describe the
characterization setup and the fuel delivery system and show results of operating the engine
while performing work. Finally, we discuss conclusions and a path toward implementing
the presented innovations in fully autonomous microrobots. This work was published in the
journal Sensors and Actuators [29].
2.2 Design and fabrication
The design of the miniature engine utilizes a cantilever SMA bending actuator [30], where
an SMA wire connects both ends of a cantilever beam. As the SMA wire is heated, it
contracts due to the shape-memory effect and as a result the cantilever beam bends. Our
8
Figure 2.4: Design of the miniature catalytic-combustion engine. (a) A segment of SMA
wires are coated with Pt. When no fuel is present to initiate catalytic combustion, the
actuator stays in its relaxed position. (b) When a gas source introduces fuel gas around the
Pt coating layer, catalytic combustion starts. The heat released from the reaction results
in the shortening of the SMA wire, which creates bending motion on the actuator. (c) The
gas source supplies an air flow to increase the cooling rate by convection. As a result, the
actuator goes back to its relaxed position.
engine consists of a carbon fiber beam with ends made of high temperature reinforced epoxy
laminate (FR4). Each end contains two orifices. An SMA wire with a length approximately
twice the beam’s length is threaded through each orifice, forming a loop and connecting
both FR4 ends. The FR4 is used to protect the epoxy present in the carbon fiber, which is
not able to stand the high temperatures reached by the SMA wire during actuation. This
configuration allows for the carbon-fiber beam to be used as a spring that holds the SMA
wires in tension, while protecting the carbon fiber from the high temperatures of the wire.
The SMA wire is coated with Platinum black powder which creates a catalytic surface.
When fuel comes in contact with the surface, a catalytic combustion reaction takes place
on the Pt coating which increases the Pt layer temperature. As a result, the SMA wire
heats up through conduction, activating the shape-memory effect and shortening its length.
The geometric constraints transform the strain generated by the SMA into a bending of
the central carbon fiber beam. When the fuel is not applied, the SMA cools down, and
the restorative force of the carbon fiber brings the actuator to its original position. Fig 2.4
illustrates a diagram with the actuator states.
Two design innovations allow us to operate this actuator at frequencies higher than 1 Hz.
First, the use of thin wires (r = 19.05 µm) decreases the thermal mass, allowing for fast
9
Figure 2.5: Fabrication of the SMA based actuators Step 1: A jig of FR4 is pre-cut using
a high resolution laser cutter. Step 2: An SMA wire is looped in the orifices of the jig and
secured at the ends using glue and knots. The SMA wire must be kept in tension. Step 3:
Using the tabs in the jig, a carbon fiber layer is glued to the back of the jig. Step 4: The
jig is aligned and cut using the high resolution laser cutter. Step 5: A thermally conductive
paste is added to the surface of the SMA. A brush is used to manually add the paste evenly
in a section of the wire. Step 6: Pt powder is poured over the actuator. The Pt powder
coats the SMA surface area with where the paste was applied. Step 7: The actuator is ready
to be used.
cooling and heating. The disadvantage of using thinner wires is that they produce smaller
forces. We overcome this disadvantage by using two wires in parallel to increase the force
generated while maintaining the fast heating and cooling. The second design innovation is
the use of the central carbon fiber, which serves two purposes. It maintains the SMA wires
in tension and it acts as a spring that helps the actuator restore its original shape when
bent.
The fabrication relies on laser cutting, manual positioning, and gluing of elements as
shown in Fig. 2.5. To cut the materials, we use a diode-pumped solid-state ultraviolet laser
(Photonics Industries DCH-355-3) with a wavelength of 355 nm and a spot diameter of 10 µm.
The process starts with a laminate of FR4 with a thickness of 127 µm. The laminate is cut
into a geometry which is used as a jig to place the actuator’s components during fabrication
(Fig. 2.5-step 1). During the laser cut, orifices are engraved on the jig which are used to
thread the SMA wires (Fig. 2.5-step 2). The SMA wire (FLEXINOL) is looped through the
orifices and tied using a simple knot to maintain its tension and secure its position. The
amount of tension is controlled visually. If the wire is not fully extended, then the tension
is insufficient. If the central carbon fiber bends due to the tension applied, then the SMA
10
is in over-tension. To prevent unraveling and to hold the wire in place in the FR4 jig, a
small amount of cyanoacrylate (CA) glue is applied carefully with a brush to the contact
area between the SMA knot and the FR4.
The carbon fiber beam is fabricated from a stack using four layers of Tenax unidirectional
prepreg carbon fiber. The fiber direction of the layers is set to [0◦ 90◦ 90◦ 0
◦
] where the 0◦
direction aligns with the bending direction. The four layers are cured via a process of pressure
and temperature to generate a final carbon stack that has a thickness of 90 µm. Rectangular
beams are then cut from the stack using the laser system, with a length and width of 10 mm
and 2 mm, respectively. After that, the carbon fiber beams are manually glued onto the back
of the FR4 jig (Fig. 2.5-Step 3). Next, a release cut is performed to separate the assembled
actuators from the jig (Fig. 2.5-step 4).
The catalytic surface over the SMA wire is obtained via a selective adhesion process.
First, thermal paste (Omegatherm 201, OMEGA) is applied to the SMA in tension via a
brush creating an area on the wire with adhesion (Fig. 2.5-Step 5). We applied the thermal
paste to create a homogeneous coating layer that covers roughly half of the SMA wire. Then,
Pt powder is poured over the actuator (Fig. 2.5-Step 6), fully covering the SMA wire. The
powder gets selectively attached to the area of the SMA wire where the paste was applied,
producing a coating surface ready for catalytic combustion (Fig. 2.5-Step 7).
2.3 Engine model
We analysed the thermal and dynamic behavior of the engine to validate the design and
predict the maximum possible frequency and displacement. A lumped capacitance heat
transfer model was used to estimate the SMA temperature during operation. The model
examines an SMA wire actuator coated with a Pt powder layer. Fig. 2.1 shows the actual
miniature catalytic-combustion engine system, while Fig. 2.1c displays the conceptual setup
of the thermal model. In this theoretical analysis, the SMA wire and the coating layer are
11
System and Dimension Bi
SMA Radial 7.2 × 10−4
SMA Longitudinal 0.38
Pt Radial 1.6 × 10−4
Pt Longitudinal 0.21
Table 2.1: Biot Number for the radial and longitudinal dimensions in the system.
treated as two concentric cylinders. The lumped capacitance assumption was used because
the Biot numbers (Bi) for the SMA wire and the coating layer were less than unity. The
Biot number was calculated using the definition in Eq. 2.1 adopted from Ref. [31]. The
summary of the results are shown in Table 2.1.
Bi =
Lcondh¯
k
(2.1)
The lumped capacitance assumption allows us to perform temporal analysis while assuming uniform spatial temperature distribution. We assume that the SMA wire can exchange
energy with the surrounding air and the coating layer, and the coating layer can exchange
energy with the surrounding air and the SMA wire (Fig. 2.1c). Additionally, energy can be
added directly to the coating layer by the heat of reaction source term Q˙
rxn. By performing
an energy balance on the SMA wire and on the coating layer, we obtain Eq. 2.2 and 2.3.
dESMA
dt = Q˙
Pt→SMA + Q˙
air→SMA (2.2)
dEPt
dt = −Q˙
Pt→SMA + Q˙
air→Pt + Q˙
rxn (2.3)
We assume constant properties of the system and constant ambient temperature (Tamb =
298 K) and relate Q˙
rxn directly to the fuel flow rate. Then we can derive Eq. 2.4 and 2.5.
12
mSMACPSMA
dTSMA
dt = USMA-PtA(TPt − TSMA)
+h¯
air-SMAA(TAmb − TSMA)
(2.4)
mPtCPPt
dTPt
dt = −USMA-PtA(TPt − TSMA)
+h¯
air-PtA(TAmb − TPt) + ∆Hrxn
˙fH2
(2.5)
The average heat transfer coefficients (h¯
air-SMA and h¯
air-Pt) were calculated using Eq. 2.6
adopted from Ref. [31]. The value of r equals the radius of the SMA wire or the outer radius
of the coating layer.
h¯ =
kair,25◦C
2r
(2.6)
The values of the heat transfer coefficient between the SMA wire and the coating layer
(USMA-Pt) were calculated using Eq. 2.7 adopted from Ref. [31]. We added β as a correction
factor that accounts for the contact imperfections between the SMA and the coating layer and
accounts for the thermal resistance of the glue layer. In our analyses, we set β = 4.9 × 10−5
.
We found that this value of β provides better agreement with experimental results, since
we observe experimentally that there are large differences between the temperatures of the
SMA wire and the coating layer.
(UA)SMA-Pt = β
ln
rPt
rSMA
2πkPtLPt
+
ln
rSMA
0.8 rSMA
2πkSMALSMA
−1
(2.7)
To deal with the reaction term ∆Hrxn
˙fH2
in Eq. 2.5, we chose Hydrogen (H2) as the
fuel and assumed that Eq. 2.8 is the only reaction path allowed. To simplify the model,
we assume that the chemical reaction time scale is much smaller than the relevant heat and
13
Figure 2.6: Plots of the Platinum coating Layer (Pt) and the SMA wire temperatures obtained from the numerical model for a specified H2 pulse train. In all simulations, the
frequency of the fuel pulse train was specified and the duty cycle was fixed at 25%. The
amplitude of the fuel pulse train was chosen to limit the maximum SMA temperature to
90◦C. As the frequency increases, there is less time available for cooling, and as a result, the
steady state temperature range decreases.
transport time scales [32]. As a result, the reaction happens instantaneously and proceeds to
completion. In the model, O2
is abundantly available. We used a constant heat of reaction
value ∆Hrxn = 1.1996 × 108J kg−1
.
2 H2 + O2
Pt
2 H2O (2.8)
The fuel flow rate term ˙fH2
(t) was modeled as an ideal rectangular pulse train with a
duty cycle of 25% and an amplitude that varies with the pulse train frequency. The duty
cycle is defined as the time fraction of the period when the fuel flow is active. The amplitude
was chosen to maximize the temperature variations of the SMA for a given frequency.
The thermal model’s system of coupled differential equations was solved numerically using
Matlab’s ode45 method. The SMA wire temperature and the coating layer temperature
14
Figure 2.7: Simulation results of the tip displacement for the miniature catalytic-combustion
engine. The displacement range decreases as the frequency of operation increases.
predicted by the model are shown in Fig. 2.6 for a given fuel flow rate amplitude and
frequency.
To evaluate the engine’s performance, we built a dynamic model of the actuator that can
predict the generated tip displacement for a given temperature profile. In order to estimate
the stress on the SMA wire in our actuator, we used a beam model with a tilt buckling
configuration adopted from Ref. [30]. Using Eq. 2.9, we can solve numerically for the
normal load on the SMA wires (P) for a given tip displacement (w).
w = b
q
P
EIL
1 − cos q
P
EIL
sin q
P
EIL
(2.9)
In Eq. 2.9, L is the effective length of the actuator, taken to be 9 mm, b is the effective
offset distance between the SMA wire and the beam neutral axis, and EI is the longitudinal
bending stiffness of the beam. We estimated EI using lamination theory to be 10.04 µN m2
[33]. The stress on the SMA can be calculated by dividing the normal load (P) by the total
cross sectional area of the two sections of the SMA wire (Eq. 2.10).
σSMA =
P
2ASMA
(2.10)
We calculated the average stress on the SMA during a cycle from the actuator’s resting
position (w = 0) to a maximum tip displacement (wmax = 2 mm) to be 145 MPa.
15
f (Hz) ∆T (
◦C) Strain Range (%) Disp. (µm)
1 26 - 90 0.00 - 2.65 2202
2 40 - 75 0.08 - 0.60 656
5 70 - 82 0.78 - 1.23 304
10 85 - 89 2.79 - 2.89 42
Table 2.2: Summary of simulation predictions for the miniature catalytic-combustion engine
at steady state operation for different frequencies.
We then calculated the SMA strain by using the Preisach-model-based description of
SMA wire dynamics described in Ref. [13]. For each frequency, we obtained the strain
profile by inputting the predicted SMA temperature profiles (Fig. 2.6) and assuming a
constant stress equal to the calculated average stress on the SMA. After, we calculated the
expected tip displacement for each strain profile using Eq. 2.11 adopted from Ref. [30]. The
obtained tip displacement (w) profiles for each frequency are shown in Fig. 2.7.
ϵSMA =
√
w2 + L2
L
− 1 (2.11)
A summary of the key steady-state results from the thermal and dynamic model are
shown in Table 2.2, where f is the frequency of the fuel flow rate pulse train, ∆T is the
minimum and maximum SMA temperatures, Strain Range is the minimum and maximum
SMA strain, and Disp. is the maximum one-way tip displacement of the actuator.
To increase the displacement of the catalytic actuator, we need to increase the strain range
of the SMA wire. For a constant stress, the SMA wire strain behaviour is correlated with
its temperature [13]. As we increase the frequency of the fuel pulse train, the SMA reaches
higher average temperatures and the difference between the high and low temperatures
within each cycle becomes smaller. This occurs because as frequency increases, there is less
time available for the SMA to reject the heat into the air. The asymmetric temperature
curve of the SMA wire, where the heating phase slope is steeper than the cooling phase
slope, indicates that to keep the SMA average temperature low, the cooling phase requires
more time than the heating phase. Additionally, as the frequency increases, the average
16
temperature of the coating layer increases, which slows down the heat rejection rate of the
SMA wire since both layers are in direct contact. There is a significant difference between
the SMA wire and the coating layer temperatures. This difference is a direct result of our
choice of β in Eq. 2.7.
Based on the simulation results and considering its assumptions, we concluded that our
proposed engine design can achieve frequencies higher than those reported in the literature
for SMA-based miniature catalytic-combustion engines. Furthermore, the dynamic model
predicts large tip displacements associated with the high operational frequencies. From the
simulations, we see that it is possible to have large differences between the average temperatures of the SMA wire and the coating layer by controlling the heat transfer coefficient
(UA)SMA-Pt. This can be achieved through insulation and through decreasing the layers’
contact area. Allowing the temperature of the SMA and the coating layer to differ widely
enables more flexibility regarding the choice of fuel. Furthermore, it is evident from the
simulation that a precise fuel mechanism is needed to deliver fuel in high frequency pulses.
Imperfections in the fuel delivery system and diffusion effects will diffuse the fuel pulses,
decrease the temperature range, and decrease the maximum displacement for a given frequency. By narrowing the width of the fuel pulses, there is more time available for cooling.
The simulation indicates that cooling is the limiting factor at higher frequencies. Thus, to
achieve higher frequencies, strategies such as active cooling must be employed.
2.4 Experiment setup
We built an experiment setup to characterize the performance of the catalytic-combustion
miniature engine and test different control strategies. We aimed to design a system that can
characterize the behaviour of different miniature catalytic-combustion SMA engines using
different gaseous fuels and different mixing ratios with air and N2
. To achieve this goal, we
built the system shown in Fig. 2.8.
17
Figure 2.8: Experimental Characterization Setup. (a.1) Gas flow sensors. (a.2) Solenoid
proportional valves connected to the sensors upstream and to the mixing junction downstream. The valves are placed on a 3D printed platform. (a.3) Laser displacement sensor.
(b) A close-up of the design of the 3D printed mixing junction and actuator base. The
mixing junction is shown in blue. The gas flows are mixed in the junction and redirected to
a nozzle that faces the actuator shown in orange. The actuator is held in place by a small
piece of tape (not shown) on one end, while the other end of the actuator is free to move
and perform work. (c) The closed loop feedback control scheme for the fuel delivery system
is shown inside the dashed red box. The overall characterization setup utilizes the displayed
open loop feedforward control scheme.
We place the actuator in the base shown in Fig. 2.8b, where one end is fixed to the base
while the other end is free. The gas flow out of the nozzle is a mixture of fuel, air, and
N2
. The flow rates and mixing ratios are controlled by the fuel delivery system. The closed
loop feedback control scheme of the fuel delivery system and the open loop feedforward
control scheme of the overall setup are shown in Fig. 2.8c. We set a reference signal for the
gas flow rates and thus control the temperature of the SMA which determines the actuator
displacement.
To obtain high frequency actuation we need a fast and precise fuel delivery system. To
achieve this goal, we utilized a custom-built fast-response high-precision fuel delivery apparatus, which consists of three air-flow sensors, three solenoid-proportional valves (Kelly Pneumatics, KPI-VP20-09025-V), pressure regulators, and a 3D printed mixing junction. For the
H2 and air gasses, we used 200 SCCM air-flow sensors (Honeywell, HAFBLF0200CAAX5).
For the N2 gas, we used a 750 SCCM air-flow sensor (Honeywell, HAFBLF0750CAAX5).
A pressure regulator was installed upstream of the flow sensors to set the maximum gas
18
Figure 2.9: Plots showing the linear displacement of the actuator tip during steady state
operation for a given fuel flow rate. The actuator is displacing a 2.15 g weight against gravity.
19
pressure allowed in the system. The different components of the fuel delivery apparatus
were connected using 1/4” tubing. The mixing juncture was 3D printed using Objet30 Pro
(Stratasys) and VeroBlue printing material. The junction was designed to minimize the
distance between the valve outlet and the SMA wire in order to minimize the diffusion of
fuel pulses.
To design the valve controller shown in Fig. 2.8c, we performed a system identification
process and designed an LTI controller using the methods detailed in Ref. [34] and [35].
The response time of the fuel delivery system to a step input when using N2 gas is 30 ms.
The flow delivery system we built is significantly faster than commercial Controlled Flow
Actuators, which have a response time in the order of seconds.
To characterize the actuation behaviour, we use a displacement laser sensor (keyence,
LK-G3001 and LK-G32). The sensor measures the linear displacement component of the
free tip of the actuator with a 0.05 µm accuracy. The experimental setup does not directly
measure the SMA wire temperature. For thin SMA wires like the ones used in this design,
most commercially available thin thermocouples are not suitable to measure the temperature
because their diameters are similar to those of the SMA wires and thus cannot produce
unintrusive measurements. Furthermore, thin thermocouples (r = 50 µm) have a maximum
bandwidth of 1 Hz and as a result cannot measure faster temperature signals [36].
All data acquisition, signal processing, and real-time control tasks are performed using
a Mathworks xPC-Target5.5 system and a PCI-6229 National Instruments AD/DA board
running at the sample-and-hold rate of 1 kHz.
The H2 reference signal was set as a pulse train with a specified frequency and a duty
cycle of 25%, similar to the fuel signal shown in Fig. 2.6. The air flow rate was chosen to
achieve a stoichiometric mixture when the H2 flow was active (the heating phase). When the
H2 flow was inactive (the cooling phase), we increased the air flow rate. Our goal of ramping
up the air flow rate during the cooling phase is to increase the convective cooling rate. The
air flow rate during the cooling phase was chosen using trial and error to get the highest
20
frequency and displacement results within the limits of the fuel delivery system. The N2 flow
was set at a constant value for each frequency. The purpose of the N2 gas was to decrease
the overall temperature of the reaction without changing the combustion equivalence ratio
during the heating phase and to help with the convective cooling during the cooling phase.
We attached a weight to the actuator’s free tip to demonstrate the engine’s ability to
perform work. In each experiment, we started with a small fuel flow rate and slowly increased
it. It usually takes the system 1 s to 10 s to reach steady state. After steady state was reached,
we increased the fuel rate again. This process was repeated until a satisfactory displacement
was observed.
2.5 Characterization results and discussion
We performed a set of experiments to investigate the engine performance while operating
at different frequencies. In Fig. 2.9, we show a summary of the engine characterization
results for 1 Hz to 6 Hz. The maximum one way displacement achieved at each frequency
is calculated and displayed in the titles of Fig. 2.9. Each figure shows the displacement
and H2 flow rate as functions of time for 6 steady state cycles. We only display the steady
state portion of the operation for clarity. For each frequency test shown, we attached a
2.15 g weight to the free tip of the actuator. As a result, the actuator is performing useful
work during each cycle. The maximum displacement achieved by the actuator is 0.9 mm
when operated at 1 Hz. We expect that the maximum displacement occurs when operating
at the lowest frequency because as the frequency increases, the time available for cooling
decreases, which increases the overall average temperature. There is a delay between the
measured fuel flow rate signal and the displacement signal. This delay is caused by the
physical distance between the valve opening and the Pt coating layer. It is also caused
by the time it takes for the heat to transfer between the Pt coating and the SMA wire.
The actuator shows significant displacements at all frequencies. The smallest displacement
21
occurred when operating the actuator at 6 Hz (60 µm).
We video recorded the actuator during all of its operational cycles. By analysing the
videos, we can verify the displacement results and learn more about the processes taking
place during the actuation cycle. In Fig. 2.10, a full cycle for the actuator is shown during
operation at 1 Hz frequency with no attached weights. The time between frames (a) and (e)
is 1 s. The combustion process taking place on the the Pt catalyst surface can be visually
observed through the red glow produced by the released heat of reaction.
The visual observations in Fig. 2.10 validate our assumption in the numerical model
that it is possible, by controlling β, to have the catalyst average temperature be significantly
higher than the SMA wire average temperature. This can be observed by noticing the glowing
red color of the Pt catalyst which indicates an approximate minimum temperature of aound
525 ◦C [37]. We also know that the SMA temperature must be below 150 ◦C, because above
that temperature the SMA is outside of its hysteresis loop and does not significantly strain
in response to temperature changes [13]. Thus, by controlling the heat transfer rate between
the SMA wire and the coating layer we can decouple their temperatures. The decoupling
allows us to use different chemical fuels than hydrogen, such as propane and butane, more
easily. Those fuels require higher ignition temperatures for the catalytic combustion to take
place [38]. The advantage of those fuels over hydrogen is their ease of storage and handling.
The decoupling acts as a low pass filter, where large oscillations in the Pt temperature lead
to small oscillations in the SMA temperature. With appropriate tuning, the Pt can be
adequately hot for the fuel to react on its surface, while the SMA is still within its operating
shape-memory effect temperatures.
We performed another set of experiments to determine the maximum work output of the
engine. In Fig. 2.11, we show frames from a video of the actuator lifting a 4.55 g mass. The
snapshot provides visual evidence of the useful work the actuator can perform. We operated
the engine with different attached weights and measures the upward displacement generated
by the engine. Then, we calculated the engine work output using Eq. 2.12.
22
WEngine = F d = mgd (2.12)
Where m is the mass of the attached weight. We measured the maximum average work
output of the engine to be 39.5 µJ at 1 Hz, corresponding to 39.5 µW of average power. This
output takes place when the attached weight is 4.55 g and for an average displacement of
0.88 mm. This result demonstrates the ability of the actuator to lift a mass 650 times its
weight utilizing catalytic combustion, corresponding to a specific power of 5.65W kg−1 when
not accounting for the fuel storage and delivery system weights.
Using the calculated engine work output, we can compute a thermal efficiency of our
proposed miniature engine using the definition in Eq. 2.13.
η =
WEngine
Fuel Energy Input
× 100 (2.13)
The fuel energy input is calculated by multiplying the total fuel input during a cycle by
the heat of reaction of H2 (∆Hrxn). The formula in Eq. 2.13 does not take into account
the energy needed to operate the fuel delivery system. The engine thermal efficiency for the
maximum average work output case was calculated to be 0.02%. The current engine design
was not optimized to have high efficiency. We suspect that the low efficiency occurs because
most of the fuel added into the system does not come into contact with the Pt layer and
thus does not contribute to heating up the SMA. To increase the thermal efficiency, we need
a more targeted fuel delivery design that decreases the amount of unreacted fuel.
The visual observations (Fig. 2.10 and Fig. 2.11) combined with the displacement
results (Fig. 2.9) confirm the simulation prediction that higher frequencies can be achieved in
miniature catalytic-combustion SMA engines. By utilizing active cooling, we can shorten the
time needed to cool down the SMA to an acceptable level to achieve significant displacement
at rates up to 6 Hz. We can achieve those frequencies by ensuring the fuel pulses are delivered
precisely and quickly. By decreasing the heat transfer rate between the catalytic coating layer
23
(a)
(b)
(c)
(d)
(e)
Figure 2.10: Frames from a video showing a full operational cycle of the catalytic actuator.
The time between frames (a) and (e) is 1 s. A red vertical line was added in the same
position in all frames to make the displacement of the actuator tip clearer. The red dot on
the actuator tip is the reflection of the displacement laser sensor. (a) The actuator is hot
and fully contracted. This is the beginning of the cooling phase. (b) As the actuator cools
down by convective and conductive forces it starts to expand, returning the actuator to its
resting position. At this point, the fuel gas reaches the catalytic layer and starts reacting.
This can be deducted from the slightly red glow appearing on the catalyst’s surface. (c)
The actuator fully reaches the relaxed position and the catalytic combustion reaction is at
its maximum rate. This is evident by the red glow seen on the catalytic layer. (d) As the
heat transfers from the catalytic layer to the SMA wire and increases its temperature, the
actuator starts to bend. Some fuel gas is still reacting on the catalyst surface. (e) The SMA
temperature is at its maximum and the actuator is fully bent and is at the beginning of the
actuation cycle again.
24
Figure 2.11: Frames from a video showing the actuator lifting a 4.55 g mass 0.91 mm at
a frequency of 1 Hz. A red line is added to signify the movement of the weight. (a) The
actuator is in its resting position. (b) The actuator is fully bent achieving its maximum
displacement and 39.5 µJ of work.
and the SMA wire we can decouple their temperatures, allowing us more operational freedom
and flexibility in terms of fuel choice.
While the simulations predicts that we can operate the actuator at 10 Hz and still observe
temperature and displacement variations, we were not able to achieve frequencies higher
than 6 Hz using the current actuator and characterization setup. Furthermore, the observed
experimental displacements were smaller than the model’s predicted displacements. The
discrepancy between the model and the experiment is caused by the simulation’s assumption
of instant fuel delivery. In the experiment, we control and measure the fuel mixture flow
rate at the gas valves. The mixture that reaches that catalytic layer differs from what we
measured because of the diffusion processes that take place in the path between the valves
and the Pt coating. Compared with the rectangular shaped fuel pulses simulated in Fig.
2.6 and measured in Fig. 2.9, the catalytic layer most likely experiences sinusoidal-like fuel
pulses with a duty cycle larger than 25%. The diffusion of the fuel pulses increases the
duration of the heating phase which does not allow for enough cooling to take place.
Additionally, the current setup is limited by the maximum speed of the valve system.
To increase the actuation frequency, we need to send narrower gas pulses in order to keep
25
the same duty cycle that allows for sufficient cooling time. Narrower fuel pulses will allow
higher frequencies comparable to those reported in the literature for electrically heated SMA
actuators [16, 21]. However, the fuel pulse width cannot be smaller than approximately two
times the response time of the fuel delivery system. In the current setup, we estimate the
minimum pulse width that we can use to be 50 µs. For comparison, if we operate the actuator
at 10 Hz and require a duty cycle of 25%, the fuel pulse width must be 25 µs. Videos of the
experimental results can be found in the supplementary materials section in [29].
2.6 Summary
We presented and tested a new design for a 7 mg miniature catalytic-combustion SMA engine that can be used for millimeter-scale robotic actuation. The innovate engine design
was made possible by using thin SMA wires combined with smart material composites. We
performed thermal and dynamic modeling to predict the maximum possible operating frequencies and associated displacements. The miniature engine was then tested on a versatile
characterization setup that can test different control strategies. To maximize the actuation
frequency, we built a custom-built fast-response high-precision delivery system. The miniature engine was demonstrated to work effectively in an operational frequency range of 1 Hz
to 6 Hz. The engine can deliver 39.5 µJ of work at a rate of 1 Hz.
The proposed miniature engine design can serve as a building block for autonomous
microrobotic applications that aim to provide onboard power using chemical fuels. The
characterization setup we built can be used to test control and fuel delivery strategies. Then,
a pre-programmed, onboard tank and valve system can deliver the fuel to the actuator in a
way that achieves the desired functional and performative requirements of the microrobots.
26
Chapter 3
Dynamics of direct propane proton exchange membrane fuel cells
3.1 Introduction
Fuel cells are electrochemical devices that directly convert chemical energy stored in fuels,
typically hydrogen, into electricity [39]. These systems have been extensively developed and
utilized in various commercial applications since the 1960s [40]. Fuel cells are commonly
categorized based on the type of electrolyte used between the anode and cathode terminals.
One common type of fuel cells is the proton exchange membrane fuel cell (PEMFC), in which
the electrolyte is a polymer membrane that conducts protons but not electrons. The typical
operation mode of a PEMFC is depicted in Figure 3.1. At the anode side, hydrogen fuel
undergoes ionization at the electrolyte surface, separating into protons and electrons. This
ionization process is facilitated by a catalyst, usually platinum-based. Protons pass through
the conductive the membrane, while electrons flow through the external circuit, generating
power. At the cathode side, electrons and protons from the hydrogen fuel combine with
oxygen molecules supplied by air streams, producing water. The overall process converts
the chemical energy of the reactants into electrical energy. The respective reactions at the
anode, cathode, and the overall cell reaction for a hydrogen PEMFC are as follows:
27
Anode reaction: H2 → 2H+ + 2e− (3.1)
Cathode reaction: 1
2
O2 + 2H+ + 2e− → H2O (3.2)
Overall reaction: H2 +
1
2
O2 → H2O (3.3)
The potential of a PEMFC for converting chemical energy into electrical energy is often
assessed by its open circuit voltage (OCV), which represents the voltage of the fuel cell when
no external current is drawn. The theoretical OCV can be determined using the formula
[39]:
OCV = −∆gf
zF
(3.4)
In this equation, ∆gf denotes the Gibbs free energy change associated with the cell’s
overall reaction, z represents the number of electrons transferred per molecule of fuel during
the reaction, and F is the Faraday constant, the charge carried by one mole of electrons.
For instance, considering a hydrogen fuel cell operating at 85 ◦C, the OCV calculation is as
follows:
OCV = −∆gf
zF
=
226, 700
2 × 96485
= 1.17 V (3.5)
This 1.17 V signifies the maximum voltage that the fuel cell can theoretically deliver under
no current load. However, practical PEMFC typically exhibit lower OCV. This discrepancy
is attributed to several factors, including activation losses, fuel crossover, and the presence of
internal currents [39]. To standardize reporting, the current and power outputs of PEM fuel
cells are typically normalized by the active area of the membrane and expressed as densities,
such as mA/cm2 or mW/cm2
.
PEMFC advantages include their low operating temperature (50 ◦C to 120 ◦C), low cost,
28
Figure 3.1: Typical operation mode of a PEM fuel cell using hydrogen fuel and air as
reactants and generating electricity, heat, and water as products.
simple construction, rapid startup, and high efficiency [41][42]. They have been used in
many large scale commercial applications such as public transportation busses in the U.S.
[43]. While there is some research activity on using methanol directly as fuel in PEMFC
[44], the majority of commercial and research PEMFC activity uses hydrogen as the fuel.
Hydrogen is the fuel of choice for PEMFC because of its high reactivity [39]. The high
reactivity of hydrogen means that large amounts of power can be generated per fuel cell
surface area (≈0.7W/cm2
). Another advantage of using hydrogen is that it produces water
and no carbon dioxide when it react with oxygen, and as a result hydrogen fuel cell system can
be considered ”zero-emission”. While the current primary source of hydrogen is natural gas
reforming, hydrogen can potentially be a clean energy storage solution. Utilizing electrolysis,
powered by renewable energy, it is possible to generate hydrogen from water, to have a “zeroemission” energy storage medium [45].
Another possible class of fuels to use in a PEMFC is hydrocarbons, namely propane and
butane. While hydrogen PEMFC are comprehensively studied and developed, direct hydro29
carbon PEMFC receive comparatively little attention and remain poorly understood. The
research presented herein is motivated by the aim to advance direct hydrocarbon PEMFC
technology for portable power generation.
In the subsequent sections of this introduction, I discuss the motivation behind my investigation, review prior research in the field, and outline the objectives of my work.
Subsequently, the rest of this chapter presents our experimental setup and methodology,
a parametric investigation of the deactivation and activation dynamics observed in direct
propane PEMFC, and the interaction between propane and other fuels, namely hydrogen
and ethylene.
3.1.1 Motivation
In this chapter, we refer to power generation in the scale of 1W to 100W as portable
power generation. For the power generation device to be considered portable, it must be
easily transported by the power consumer. The typical application for this type of power
generation we are interested in is portable consumer electronics such as laptops. The vast
majority of portable power generation today utilizes chemical batteries, namely lithium-ion
batteries [46].
While they have many advantages, most batteries, including lithium-ion batteries, have
significantly lower specific energy than chemical fuels, such as hydrogen and hydrocarbons.
For example, state of the art lithium-ion batteries have specific energy of 0.875 MJ kg−1
,
compared to 141.80 MJ kg−1
for hydrogen and 50.35 MJ kg−1
for propane [17][47]. The power
density of state of the art lithium-ion batteries is 340W kg−1
. Determining the power density
of hydrogen and hydrocarbons is more difficult since the power densities will depend on the
conversion device used to convert the chemical energy in the fuels to mechanical or electrical
energy. The choice of conversion device is highly dependant on the scale of power required
and the application specifics.
Given the significantly higher energy density of hydrogen and propane, over 50 times that
30
of batteries, it is both logical and desirable to explore portable power generation devices
that utilize chemical fuels. Furthermore, portable devices powered by chemical fuels can
potentially offer additional advantages, including instant rechargeability, reduced use of toxic
materials, absence of self-discharge, and elimination of memory effect.
To illustrate the advantages of portable power generation devices that utilize chemical
fuels, consider a PEMFC system compared to a battery system. Operating a PEMFC at 10%
efficiency, the fuel cell system has the potential to be approximately 5 times more energydense than batteries, given that hydrogen and propane possess energy densities more than
50 times higher than that of batteries. Although the overall weight of the fuel cell system,
including the fuel storage apparatus, reduces this advantage, PEMFC could still offer higher
energy density compared to a battery system. In practice, hydrogen PEMFC systems has
considerably higher efficiencies than 10%, typically around 70% [48]. Moreover, PEMFC,
due to their operational characteristics at low temperatures, emerge as a suitable choice for
portable power applications among other fuel cell types.
Although hydrogen is the standard fuel for PEMFC, it is not practical at portable scales
due to the challenges associated with its storage and handling. Storing hydrogen poses significant challenges at any scale, including concerns related to safety, material compatibility,
and cost-effectiveness [45]. At small scales, these challenges are exacerbated, as the specific
energy and energy density of hydrogen is significantly reduced when factoring in the weight
of the fuel storage apparatus. Alternatively, direct methanol PEMFC can be used to mitigate the storage and handling issues of hydrogen. However, methanol PEMFC suffer from
fuel crossover [49]. Another disadvantage of using hydrogen or methanol is that they are
currently both obtained through complex reforming processes of natural gas, and do not
occur naturally.
Based on those challenges, a direct hydrocarbon PEMFC is a desirable alternative for
portable power generation because it can mitigate the storage and handling issues of hydrogen
PEMFC and the crossover and fuel availability limitations of direct methanol PEMFC.
31
We envision that with significant research investment, a portable, direct propane or
butane PEMFC can be designed and used to power portable electronics. Because of this
potential for direct propane PEMFC, we pursued this project to investigate their dynamics,
and devise ways to improve their performance.
3.1.2 Previous Work
Researchers started exploring using hydrocarbons and other chemical fuels as a fuel source
in fuel cells since the 1960s. Niedrach used a fuel cell with a sulfonated phenolformaldehyde
acid electrolyte to demonstrate propane direct anodic oxidation producing 1.5 mW/cm2 at
85 ◦C with 18 mg/cm2 of platinum catalyst loading [50]. He found that platinum performed
better than palladium. Grubb et. al. demonstrated a propane fuel cell using concentrated
phosphoric acid electrolyte, platinum black catalyst, and operating at 200 ◦C to produce
50 mW/cm2 at 0.5 V [51]. Cairns demonstrated a propane fuel cell using aqueous cesium
salt, platinum black catalyst, and operating at 150 ◦C to produce 80 mW/cm2
. [52]. In all
those studies, authors observed that hydrogen performed significantly better in the same
cells.
The first work we are aware of that reports a working propane fuel cells using modern
polymer membranes as an electrolyte was done by Savagado and Rodriguez. They investigated the performance of direct propane PEMFC using Nafion N-117 membranes as electrolyte and different platinum based catalysts operating at a temperature of 80 ◦C. For 20%
PtOx/C anode catalyst they obtained maximum power density of 14.5 mW/cm2
[53]. W.S.
Li et al. investigated ethane, propane, and butane direct fuel cells using Nafion N-117 membrane and carbon-supported platinum [54]. He reported power densities of 1.47 mW/cm2
for
propane. He also concluded that there is a positive correlation between the power density
and the temperature and the catalyst loading.
32
The overall reaction in a direct propane PEMFC is
C3H8 + 5O2 → 3CO2 + 4H2O (3.6)
While the anode and cathode reactions, respectively, are
C3H8 + 6H2O → 3CO2 + 20H+ + 20e− (3.7)
5O2 + 20H+ + 20e− → 10H2O (3.8)
The theoretical OCV for a direct propane fuel cell operating at 85 ◦C can be calculated
using Eq. 3.4 as follow:
OCV = −∆gf
zF
=
2, 136, 000
6 × 96485
= 1.11 V (3.9)
The first step of the reaction in the anode is expected to be the dehydrogenating of
propane and water, given by
C3H8 → C3H7 + H (3.10)
H2O → H + OH (3.11)
H → H
+ + e− (3.12)
Which demonstrates the importance of water presence in the anode [55]. Note that the
produced electron travels through the external circuit, while the proton is conducted through
electrolyte membrane.
Researchers have also utilized numerical simulations to investigate various aspects of direct propane PEMFC. Vafaeyan et. al. numerically examined the performance of nickel
alloy catalysts in the anodes of high temperature direct propane PEMFC [55]. This study
explored how different catalyst compositions could mitigate challenges associated with cat33
alyst poisoning, adsorption, and activation energy. Further modeling work was conducted
by Khakdaman et. al., employing computational tools to assess the impact of operational
parameters, such as temperature, propane concentration, and flow rate, on the performance
and efficiency of direct propane fuel cells [56].
Recently, E. Antolini published a comprehensive review on the recent developments and
future directions of research in direct propane fuel cells, including PEMFC [57]. This work
highlighted the potential for developing standalone portable propane fuel cells.
3.1.3 Summary of previous work in our lab
Based on our motivation and the previous work reported in the field, my research group, the
Combustion Physics Lab at the University of Southern California under the supervision of
Dr. Paul David Ronney, started a multi-year effort to investigate direct propane PEMFC.
This work was done in collaboration with Dr. G. K. Surya Prakash.
The initial phase of this study was led by Dr. Eugene Kong in collaboration with myself
[58][59]. The objective was to replicate abd build upon the findings reported by Savagado
and Rodriguez (2006). Kong et. al. used a direct propane PEMFC with a platinum black
catalyst and operating at a temperatures of 80 ◦C. High catalyst loading was typically
employed in this work, 8 mg/cm2
to 15 mg/cm2
. When they fueled the cell with research
grade propane, which has a purity of 99.99%, the cell open circuit voltage (OCV) remained
below 0.05 V, which indicates very low potential difference between the reactants. This was
described as the cell’s failure to “ignite”, drawing an analogy to combustion terminology.
Since Savagado and Rodriguez did not report the purity of the propane fuel they used,
Kong systematically investigated the impact of introducing common impurities typically
present in less pure propane into his fuel stream. They found that the addition of certain
unsaturated hydrocarbons, such as ethylene, resulted in an open-circuit voltage (OCV) of
0.8 V, enabling power generation. Remarkably, the influence of these additives persisted even
when they were only added to the fuel stream for short periods. For instance, introducing
34
2000 parts per million (PPM) of ethylene into the propane fuel stream for a duration as short
as 5 s facilitated sustained cell operation, producing power over extended durations exceeding
1000 s. A positive correlation between power density and impurity additive concentration
was observed. Using impurity additives, Kong et al. were able to reproduce Savagado and
Rodriguez’s results, suggesting the latter utilized lower purity propane in their work.
Moreover, Kong et. al. observed dynamic behavior in a direct hydrocarbon PEMFC
fueled by research-grade propane. Conventionally, a hydrogen PEMFC operates at a specific
current and voltage point. Initially, when no current is applied to the cell, it operates at the
OCV point. Subsequently, upon the application of a predetermined current load, the voltage
decreases, stabilizes over time, and the cell can then operate nearly indefinitely, provided
an adequate supply of reactants and appropriate humidity and thermal management. For
a given set of operational conditions, a hydrogen PEMFC typically does not demonstrate
significant current, voltage, or power dynamics.
In contrast, Kong et. al. observed pronounced power dynamics in a direct propane
PEMFC, strongly influenced by current densities and fuel purity, as shown in Figure 3.2.
When subjecting the cell to a constant current load, a progressive decline in cell voltage occurs over time, consequently diminishing the power output. This voltage decay accelerates
with time, culminating in the abrupt cessation of power generation, a phenomenon termed
“cell extinction” by Kong et. al. This dynamic behavior is dependant on both the applied
current density and the concentration of impurities introduced. For lower impurity concentrations, cell extinction occurs within approximately 100 s. Conversely, at higher impurity
concentrations, the onset of extinction is delayed, surpassing 1000 s and potentially persisting
indefinitely.
Kong et al. hypothesised that the phenomenon of cell ’extinction’, and the relatively low
power densities of propane PEMFC, stems from a catalyst deactivation mechanism, which
turns active reaction sites inactive. They observed that discontinuing the current load on the
cell, followed by its reapplication, reactivates the cell and resets dynamics. This observation is
35
30
1. It is easier to oxidize UHs compared to hydrocarbons and since UHs oxidation is an
exothermic reaction, it “triggers” propane oxidation.
2. Some species might be adsorbed on the platinum surface and UHs are helping to
remove those species from the platinum surface.
3. Ethylene might be reacting with oxygen (crossover from cathode and oxygen is fatal to
hydrocarbon PEMFCs) to form ethanol or DME, which are much easier to oxidize than
that of propane.
Figure 3.2. Effect of ethylene concentration on power production at constant current 36 mA/cm2
(for 0 ppm, the cell was first “ignited” with ethylene then the ethylene flow was stopped). Note
logarithmic time scale.
Figure 3.2: Effect of ethylene impurity concentration on power production at constant current
of 36 mA/cm2
(for 0 PPM, the cell was first “ignited” with ethylene then the ethylene flow
was stopped). Note the logarithmic time scale [58].
illustrated in Figure 3.3, where they employed a “load-interrupt” mode, alternating between
the application and cessation of current load to repeatedly reset the cell dynamics.
Furthermore, Kong et al. explored the use of butane as a direct fuel in PEMFC and
found that both propane and butane exhibited essentially the same deactivation behavior
and achieved comparable power densities. While our study concentrates on direct propane
PEMFC, we anticipate that these results would extend to direct butane PEMFC as well.
Throughout the remainder of this study, we designate the phenomenon of voltage and
power decline over time as “deactivation”, while the voltage recovery process of the cell is
termed “activation” dynamics.
Building upon our initial research directed by Dr. Kong, I proceeded with the investigation into the activation and deactivation dynamics of the direct propane PEMFC. The
overarching objective of this research is to improve our understanding of the underlying
mechanisms, with the ultimate aim of making direct propane PEMFC viable for portable
power applications.
36
38
Figure 3.7. Constant current and load-interrupt mode of a 25cm2 direct propane PEMFCs. The
flow rate was 1.2 L/min and 0.8 L/min for propane and oxygen, respectively. The operating
temperature was held at 80˚C. The current was applied for 20 seconds and shut off for 5 seconds.
0
2
4
6
8
10
12
14
16
0 50 100 150 200 250 300 350 400 450 500
Power!Density!(mW/cm2)
Time!(Seconds)
LoadFInterrupt:!12!mA!cm^F2 Constant!Current:!12!mA!cm^F2
0
5
10
15
20
25
30
0 50 100 150 200 250 300 350 400 450 500
Power!Density!(mW/cm2)
TIme!(Seconds)
LoadFInterrupt:!24!mA!cm^F2 Constant!Current:!24!mA!cm^F2
Figure 3.3: Constant current (galvanostatic) and “On-Off” modes of a 25 cm2 direct propane
PEMFC. The flow rate was 1.2 L/min and 0.8 L/min for propane and oxygen, respectively.
The cell temperature was held at 80 ◦C. For the On-Off mode, the current was applied for
20 seconds and then shut off for 5 seconds, repeatedly [58].
3.1.4 Objectives
The underlying goal of this work is to understand the mechanisms behind the observed
dynamics of low-temperature direct propane PEMFC to design and operate cells with reasonable power densities that compete with portable batteries. The project has three main
objectives: first, to build a testing apparatus that can be used to investigate direct propane
PEMFC; second, to develop a deep understanding of the cell dynamics; and third, to use
that understanding to devise effective operational strategies.
To improve our understanding of the dynamics, we aim to systematically study the effect
of operational factors on the deactivation and activation dynamics. These factors include
current density, fuel flow rate, catalyst loading, membrane electrode assembly type, cell
temperature, and cell state before reactivation. The motivation for this study is that by
understanding how the dynamics depend on these operational factors, we can potentially
exploit the dynamics to improve the performance of the cell.
To devise effective operational strategies, we will model the cell dynamics based on the observed experimental results. Then, using this model, we will design and test different control
37
strategies with the aim of improving the cell’s performance and power output. The testing
of the control strategies will be conducted both in simulation and through experiments.
3.2 Experiment methodology
3.2.1 Experiment apparatus
To investigate the dynamics of direct propane PEMFC experimentally, a custom-built testing
apparatus was constructed. The previous setup utilized by Kong et al. [59] was unavailable,
necessitating the construction of a new setup tailored to the research objectives. Emphasis
was placed on ensuring flexibility in the setup design to accommodate various operational
arrangements. This flexibility was deemed essential, as commercially available all-in-one
testing systems typically offer limited customization options. A detailed description of the
setup is provided below to allow for replication by other researchers.
Furthermore, in addition to flexibility, our custom-built apparatus offers significant cost
savings compared to commercially available all-in-one testing setups commonly found in fuel
cell laboratories. By sourcing and assembling each component individually, we estimate that
our setup cost approximately $8,000, whereas a comparable system from Scribner Inc. is
priced upwards of $30,000 [60].
A conceptual diagram of the setup is shown in figure 3.4. The completed physical setup
is shown in Figure 3.5 and it consists of the following main components:
1. Mass flow controllers
2. Heated humidifiers
3. Electronic load device
4. Thermocouples and data acquisition systems
5. Fuel cell assembly
38
Fuel Cell
W/ Temp control
Mass Flow
Controllers
Humidifier
W/ Temp control
Humidifier
W/ Temp control
Electronic Load
Computer Control
Software
Water Water
Oxygen
Fuel
Figure 3.4: Conceptual diagram of the fuel cell testing apparatus we built.
6. Computer control software
To regulate the gas flow into the cells, five TELEDYNE VUE 300 digital mass flow
controllers (MFCs) were employed. The output of one pair of MFCs was combined and
directed into one of the humidifiers, while the output of the other pair was similarly combined
and directed into the other humidifier. The remaining MFC’s output was directly channeled
into the PEMFC’s anode flow stream. This latter MFC possessed a had capacity and was
exclusively utilized when incorporating additives into the fuel flow stream at flow rates below
20 SCCM. All MFCs were computer-controlled, as detailed in the subsequent discussion.
For gas humidification prior to entry into the anode and cathode sides of the fuel cell, a
stainless steel bubbler from LabCommerce was employed. The bubbler has 1 L capacity and
a 1 µm sintered stainless steel element welded to the end of the dip tube. A heating element
was fitted around the humidifier to regulate the water temperature, thereby controlling the
temperature of the gases flowing into the cell. Prior to each experiment, the humidifiers were
39
Experimental setup
Flow Control
Electronic Load
Humidifiers + heaters
Heater controller
DAQ
Fuel Cell (MEA inside)
Figure 3.5: Picture of the main components of the fuel cell testing apparatus we built in the
lab.
refilled with deionized water to full capacity. The temperature of the humidified gases was
maintained at 5 ◦C above the fuel cell temperature to ensure that the gases achieved 100%
relative humidity upon entering the cell. The humidified heated gasses are directed into the
cell in 0.25 in thermally insulated metal tubes to the fuel cell assembly.
The electronic load applied on the cell was obtained using a galvanostat from multicomPRO (MP710258US) that can run in constant-current or constant-voltage modes and handle
up to 30 A and 150W. The galvanostat was used to apply the current and read the voltage
across the cell. The galvanostat was controlled using the computer control software to run
different testing programs and record the data.
Temperature regulation of both the fuel cell and the humidifiers was accomplished using
multiple type-K thermocouples. The temperature of the humidifiers was measured on the
exterior surface of the bubbler. Validation experiments were conducted to establish a 1:1
correlation between the surface temperature of the humidifier and the temperature of the
gas prior to entering the fuel cell. The temperature within the cell was monitored via a
40
dedicated orifice within the fuel cell assembly, facilitating temperature measurement at the
midpoint of the cell’s graphite separator plates. Data acquisition from the thermocouples
and control signals for the various heaters were managed by National Instruments DAQ
boards (NI USB-6001), connected to the computer control software..
The fuel cell assembly is made up of the following components:
1. Current collector plates
2. Graphite separator plates
3. Teflon gasket
4. Gas diffusion layer
5. Catalyst layer
6. Polymer electrode membrane
The design is based on fuel cell assembly used by Kong et. al. as shown in Figure
3.6 [58]. The current collector plates and graphite separator plates were purchased from
the FuelCellStore made for a 25 cm2 membrane active area. The current plates are made
of copper plated with gold and are connected to the external load. The graphite plates
conduct the current to the current collector plates. The Teflon gaskets are made from
PolyTetraFluoroEthylene film made with Teflon fluoropolymer with thickness of 0.005 in.
The gaskets prevents leaks from between the anode and the cathode halves. Two silicon
adhesive heating pads are attached to the outside of the current collector plates. The heating
pads ensures the cell is operating at the desired temperature. The gas diffusion layer, the
catalyst layer, and the polymer electrode membrane are discussed in section 3.2.2.
To control all aspects of the experiment, a comprehensive program was created in LabVIEW [61]. The program controls the gas flowrates, temperatures of humidifiers and the
cell, and the electronic load. Additionally, the program handles all data acquisition and
recording. Because the LabVIEW program was custom-made, it allows us the flexibility to
41
22
Figure 2.1. Schematic drawing of a single fuel cell apparatus 1) current collector plates plated with
gold for durability 2) graphite flow fields 3) gaskets to prevent gas or liquid leakage 4) carbon
paper for gas diffusion layer to enhance diffusion of reactants 5) platinum black for electrooxidation of propane 6) PEM for proton conductivity [27].
In the order shown in figure 2.1, each components were assembled together and held in
place with eight long bolts and nuts. In order to prevent electrons flowing from anode to cathode
via metal bolts, all bolts are equipped with plastic sleeves on both ends.
Figure 3.6: Schematic drawing of a single fuel cell assembly apparatus [58].
Name MEA PtB CPL MEA PtB FCS MEA PtC FCS
Maker Made in CPL Lab FuelCellStore FuelCellStore
Catalyst type Pt Black Pt Black 60wt% Pt on Vulcan
Catalyst loading 8 -12 mg/cm2 4 mg/cm2 0.5 - 1.5 mg/cm2
Catalyst application Ink and Brush Proprietary Proprietary
Membrane Nafion N-117 Nafion N-212 Nafion N-212
Gas diffusion layer Carbon paper Woven Carbon W1S1011 Woven Carbon W1S1011
Active Area 5 x 5 cm2 5 x 5 cm2 5 x 5 cm2
Table 3.1: Comparison of the different MEAs used in the experiments.
conduct standard fuel cell tests such as constant-current tests and polarization curve tests,
as well as non-standard operation modes that will be discussed later. A screenshot of the
LabVIEW program UI is shown in Figure 3.7.
3.2.2 Membrane electrode assembly (MEA)
The gas diffusion layer, the catalyst layer, and the polymer electrode membrane together are
referred to as the membrane electrode assembly (MEA). We have used three types of MEAs
in our work. Table. 3.1 summarizes the key differences between them. In the following
chapters, we will use the MEA name stated in the table to refer to the different types of
MEAs.
42
Unified Controller v.5.0.vi
G:\My Drive\USC\CPL\Setup\Unified controller\Unified Controller v.5.0.vi
Last modified on 3/19/2024 at 3:48 PM
Printed on 3/19/2024 at 3:53 PM
Page 1
0.0000V 0.0000V 0.0000V Voltage Measured (V) 0.0000A Current Measured (A) COM7 VISA Refnum in
50 while loop delay ms 15 delay between each reading (ms) 0 Manual Set Current (A)
00:01:22
time (s)
LOAD ON/OFF Switch
0.0584371
Set Current Value (A) Save Data? File_name notes Current Input Options / Tests
General Settings E-Load Controls Data Save E-Load Flow Box H2
MFC 1
O2
MFC 2
N2
MFC 3
N2
MFC 4
N2
MFC 5 MASTER Flow control ON/OFF
ID
0.1
Setpoint (in SCCM) 0 Measurement
Initiate? Initiate?
ID 2
0.1
Setpoint (in SLM) 2
0 Measurement 2
Initiate? Initiate? Initiate?
ID 3
0.1
Setpoint (in SLM) 3
0 Measurement 3
ID 4
0.1
Setpoint (in SLM) 4
0 Measurement 4
0.1
Setpoint (in SCCM) 5
0 Measurement 5
3
-0.1
0.5
1
1.5
2
2.5
Time 42:56 44:10 45:00 45:50 46:40 47:30 48:20 49:10 50:25 MFC 1 MFC 2 MFC 3 MFC 4 MFC 5 Waveform Chart
-0.0295
voltage [V] -2.92
Temp [c]
Thermal Readings Fuel Cell Thermo-couple and control 85
setpoint cell temp
0.1 Hysteresis, h Cell Heater cell heater master switch
115
20
40
60
80
100
Time 0 200 400 600 800 1023 Waveform Chart 3 Plot 0 MPC
Current Mode / Test Case Start Test
25 Cell Size (cm2)
2
Scan Rate (mA/s cm2)) 0Max Current Density mA/cm2 2
Stabalization time (s) 0max current A
00:00:00
total test time 00:00:00
time left
0Power (mW) 25 Cell Size 0Power density (mW/cm2) 0
current density (mA/cm2)
1.2
0.0
0.2
0.4
0.6
0.8
1.0 Current (A) -1.00 -0.80 -0.60 -0.40 -0.20 0.00 0.20 0.40 0.60 0.80 1.00 XY Graph 2 Plot 0 Reset 1.0
-1.0
-0.8
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
0.8 Current (A) -1.00 -0.80 -0.60 -0.40 -0.20 0.00 0.20 0.40 0.60 0.80 1.00 XY Graph Plot 0
0
Temp offset +/- [C]
0
fuel stream HHV (MJ/kg) 0
fuel molar mass (g/mol) 0max theoritical power (W) NaN
efficiency (%) Polarization Curve mode - settings Manual CC Mode 0Manual Set Current Density (mA/cm2) 25 Cell Size 0Current (A) 25 Cell Size 0
Experiment ID # On Off mode - settings 0Off time (s) 0On time (s) 0 Off Curr d (mA/cm2) 0On Curr d (mA/cm2) 0Off Curr (A) 0On Curr (A) 0.1
temp update wt 0GCF 1
0GCF 2
0GCF 3
0GCF 4
0GCF 5
82
-1
10
20
30
40
50
60
70
Time 388 400 420 440 460 480 493
current density (mA/cm2) vs time 20
0
2.5
5
7.5
10
12.5
15
17.5
Time 387 400 410 420 430 440 450 460 470 480 493 power density (mW/cm2) vs time 0.9
0
0.2
0.4
0.6
0.8
Time 0 200 400 600 800 1023
voltage vs time
0.008 Capacity [slm] 0measurment (SCCM) 0
voltage in
0Additive ppm
0
voltage out
1273708001
ID 1
0790074003
ID 2
0790086001
ID 3
0790101002
ID 4
ID 5
serial#
COM1
COM7
0
all COM ports used none none none none none 0 corres. serial # (check)
0 corresponding COM
com1 com2 com3 com4 1, fuel2 2, Air2 3, Air1 4, fuel1
0
additive ppm
Controller V1
0
I_h
0
I_l 0V_l 0V_h
0branch
0mean Power density (mW/cm2)
8
0
1
2
3
4
5
6
7
Time 387 400 410 420 430 440 450 460 470 480 493 mean power density (mW/cm2) vs time
0.1 power den avg wt
1
volt stab wt 50.0
-5.0
0.0
5.0
10.0
15.0
20.0
25.0
30.0
35.0
40.0
45.0 Voltage (V) -1.00 -0.75 -0.50 -0.25 0.00 0.25 0.50 0.75 1.00 XY Graph 3 Plot 0
flush
200 N_flush N 493
MPC
400
I_max 10 p
50 dIdt_max 0.5 dt_MPC
Figure 3.7: Screenshot of the LabVIEW UI used to perform fuel cell testing.
43
We extensively employed commercial membranes sourced from FuelCellStore [62]. The
primary advantage of these membranes is that they are easily obtainable and that the supplier asserts that their catalyst application methodology ensures uniformity and repeatability.
However, the supplier provides limited details regarding the catalyst application process and
quality assurance procedures. The commercial membranes utilized in our study incorporated either Pt black catalyst (PtB) or 60wt% Pt supported on Vulcan carbon (PtC). The
maximum catalyst loading offered by FuelCellStore was 1.5 mg/cm2
for PtC and 4 mg/cm2
for PtB. Recent trends in fuel cell applications have favored using supported Pt catalysts
due to their comparable performance to Pt black while requiring lower Pt quantities, thus
reducing costs. The commercial membranes we used utilized Woven carbon cloth W1S1011
as the gas diffusion layer. The supplier does not disclose information regarding the assembly
process of their MEAs or the associated conditions such as temperature and pressure.
In addition to utilizing commercial membranes, we fabricated our own membranes in
the laboratory following the methodology outlined in [59]. Specifically, a Pt catalyst ink
was prepared made of 1 part platinum black, 20 parts deionized water, 2 parts Nafion resin
solution [54], and 6 parts ethanol, by weight. The ink was thoroughly mixed via stirring
before being manually applied to a 25 cm2
carbon paper using a paintbrush. Examples of
the carbon paper with PtB catalyst applied and subsequent drying are shown in Figure 3.8.
Catalyst loading was determined by weighing the carbon paper before and after catalyst
application. Two carbon papers were then combined against a Nafion membrane and hotpressed with a total force of 2100 N at a temperature of 120 ◦C for 20 minutes to form the
MEA.
The in-house fabricated membranes offered the advantage of high catalyst loading, with
loadings reaching up to 13 mg/cm2 PtB utilizing the painting technique. However, the
drawback of this approach lies in the non-uniform application of platinum and the inability
to precisely control loading amounts. This is evident in Figure 3.8 in the uneven distribution
of catalyst (appearing in in dark black) on the carbon paper (appearing in gray). As a
44
Figure 3.8: The carbon paper with catalyst painted on it after it dried and before it is pressed
onto the Nafion membrane to form MEA PtB CPL . The carbon paper on the left had a
loading of 12.2 mg/cm2 and the carbon paper on the right had a loading of 9.2 mg/cm2
.
result, the performance of the final MEA is heavily reliant on the technique of the researcher
fabricating the MEA, resulting in variability and decreased experiment repeatability. The
MEA PtB CPL used in this work had anode and cathode catalyst loadings of 12.2 mg/cm2
and 9.2 mg/cm2
, respectively.
3.2.3 Cell operation
For the propane experiments, research grade propane was used (99.99% purity, PROPRESLP5).
This is the same grade used by Kong et. al. [59]. Gas chromatography analysis (Agilent 8890
GC System) was used to verify the propane purity. Unless otherwise specified, the anode
propane flow rate was set to 1.2 L/min, the oxygen cathode flow rate was set to 0.8 L/min,
and the cell temperature was set to 85 ◦C. The humidifier internal temperature was always
set to 5 degrees higher than the cell temperature to ensure 100% relative humidity of the
reactants entering the fuel cell.
To initiate fuel cell operation, the system was first purged with heated, humidified nitrogen on both the anode and cathode sides for a duration of 90 minutes. After the cell
reached the desired temperature and was adequately purged and humidified, the flow of fuel
and oxygen reactants was initiated, and the nitrogen flow was ceased.
45
3.3 OCV with research grade propane
The open circuit voltage (OCV) in a fuel cell serves as a crucial parameter indicative of
its performance and operational state. Essentially, the OCV represents the voltage output
of the fuel cell when no electrical load is connected, offering insights into its electrochemical potential. Interpreting the OCV involves understanding the balance between the cell’s
electrochemical reactions, including fuel oxidation and oxygen reduction, and the associated
thermodynamic driving forces. A higher OCV typically signifies a more favorable balance of
these reactions, indicating enhanced cell efficiency and power output potential. Conversely,
a lower OCV may suggest inefficiencies or limitations within the cell’s operation [39]. Furthermore, changes in the OCV over time or under varying operating conditions can provide
valuable information about the cell’s stability, kinetics, and overall performance. Thus, monitoring and analyzing the OCV play a pivotal role in optimizing fuel cell design, operation,
and diagnostic procedures.
Using the newly constructed experimental setup, we initiated efforts to replicate the
results obtained by Kong et al. Utilizing all three types of MEAs as detailed in Table 3.1, we
achieved a nonzero OCV and generated power using only research-grade propane (99.99%
purity), without the addition of any impurity additives to the propane fuel flow. Figures
3.9, 3.10, and 3.11 shows the observed OCV dynamics for each MEA type.
For all membrane electrode assemblies (MEAs), the open circuit voltage (OCV) exhibits
a gradual increase as fuel flow is directed into the cell. Figure 3.9 illustrates that this increase
in OCV occurs at a slower rate for lower flow rates. It is noteworthy that, unless otherwise
specified, a propane flow rate of 1.2 SLM was utilized in the experiments reported herein.
In all MEAs, the OCV reaches its maximum value within the timeframe of approximately
100 s to 200 s, following which it gradually decreases until reaching a stabilized level. The
stable OCV typically corresponds to around 75% of the peak OCV value.
For both MEA PtC FCS and MEA PtB FCS, the stabilized OCV was observed to con46
0 20 40 60 80 100 120 140 160
Time (s)
0.1
0.2
0.3
0.4
0.5
0.6
V
olta
g
e (V)
0.1 SLM
0.5 SLM
0.8 SLM
1.2 SLM
Figure 3.9: MEA PtC FCS OCV dynamics for different propane flow rates with 0.5 mg/cm2
catalyst loading. The fuel flow was started at time 0 and the voltage was measured until it
stabilized. The OCV was observed to be stable for more than 3000 s. MEA PtC FCS with
1 mg/cm2 and 1.5 mg/cm2
catalyst loadings exhibited similar OCV dynamics and stability.
0 200 400 600 800 1000
Time (s)
0.0
0.1
0.2
0.3
0.4
V
olta
g
e (V)
Figure 3.10: MEA PtB FCS OCV dynamics with 4 mg/cm2
catalyst loading. The fuel flow
was started at time 0 and the voltage was measured until it stabilized. The observed voltage
oscillations were confirmed to correspond to the fuel cell temperature oscillations, which were
caused the on-off cell temperature controller’s non-zero dead-band.
47
0 250 500 750 1000 1250 1500 1750 2000
Time (s)
0.2
0.4
0.6
0.8
V
olta
g
e (V)
Figure 3.11: MEA PtB CPL open circuit voltage (OCV) dynamics with 12.2 mg/cm2 and
9.2 mg/cm2
catalyst loading on the anode and cathode sides, respectively.The fuel flow was
started at time 0 and the voltage was measured until it stabilized.
verge at approximately 0.3 V, whereas for MEA PtB CPL, this value was around 0.7 V. This
discrepancy may be attributed to differences in catalyst loading. The MEA PtC FCS tested
had a loading ranging from 0.5 mg/cm2
to 1.5 mg/cm2
, which aligns with typical values for
hydrogen PEMFC but may be considered low for direct propane PEMFC, given propane’s
lower reactivity compared to hydrogen within the cell. Similarly, the MEA PtB FCS had
a loading of 4 mg/cm2
, typical for hydrogen PEMFC but potentially insufficient for direct
propane PEMFC. The catalyst loading could not be increased beyond these levels due to
technical limitations from the MEA commercial supplier. The MEA PtB CPL we fabricated
in house, had significantly higher loading, 12.2 mg/cm2 and 9.2 mg/cm2 on the anode and
cathode sides, respectively. This loading is much higher than loading typically used in hydrogen PEMFC. The higher loading might explain why MEA PtB CPL had a significantly
higher OCV.
We have also found that the OCV is significantly influenced by the cell temperature, as
shown in Figure 3.12 for the MEA PtB CPL. The OCV increased with temperature, peaking
at approximately 85 ◦C. Beyond this point, further temperature increases did not increase
the OCV. At lower temperatures, the OCV was observed to be less than 0.4 V. A noticeable
transition in the OCV occurred within the temperature range of 40 ◦C to 70 ◦C. Testing at
48
20 30 40 50 60 70 80 90 100
Fuel Cell Temperature (C)
0.0
0.2
0.4
0.6
0.8
1.0
Open Circuit Voltage (V)
Figure 3.12: MEA PtB CPL open circuit voltage (OCV) as a function of cell temperature.
higher temperatures was not possible due to hardware limitations.
Our OCV findings align with those reported by Savagado and Rodriguez in [53], despite
their use of different platinum-based catalysts. Savagado and Rodriguez observed OCV
values around 0.8 V under similar conditions for a direct propane PEMFC. Although they
did not specify the propane’s purity, it is believed to be lower than research-grade propane,
considering our previous discussion. In our study, the MEA PtB CPL achieved a peak OCV
of 0.9 V and maintained a stable OCV at 0.7 V, utilizing only research-grade propane with
a significantly high catalyst loading.
Contrary to Kong et al.’s findings in [59], we achieved a nonzero OCV using researchgrade propane without any fuel additives for all MEA types tested. Additionally, our cells
generated significant power densities with research-grade propane alone, results we’ll detail
later. Kong et al. noted achieving OCV values up to 0.8 V by “igniting” the cells with
unsaturated hydrocarbons like ethylene, using MEAs with high catalyst loading similar to our
MEA PtB CPL. They also observed that the additive’s effect lasted well beyond its removal,
sometimes exceeding 1000 s. Our results, obtained with just research-grade propane, mirror
Kong’s post-additive results. This leads us to suspect that the difference in our results,
especially related to OCV and Kong’s inability to “ignite” his cell, might be due to testing
hardware differences.
49
The Scribner system used by Kong is mainly used to test hydrogen PEMFC, and it
might have been unable to detect the low voltage and low powers produced by the direct
propane PEMFC. When the additive is added, the voltage and power spike initially, which
might cause the system to detect that signal and then persist in detecting it. Using our
testing setup, we were unable to replicate the “no ignition” findings of Kong et. al under
any condition.
Our ability to achieve a non-zero OCV and use it to generate power is significant because
it allows us to study the dynamics of direct propane PEMFC, using high purity research
grade propane alone. This allows us to eliminate the effects of other fuel impurities and
might shed a light on the deactivation mechanisms taking place in the cell.
3.4 Deactivation dynamics
Similar to Kong et al.’s findings, our direct propane PEMFC experienced deactivation dynamics. Deactivation here refers to the phenomenon where the cell’s voltage decreases under
a constant current load until it drops to very low values (typically below 0.05 V), leading to
a reduction in the cell’s power output to zero. This deactivation, observable by monitoring
the cell voltage or power density. The power density is related to the cell voltage by Eq.
3.13.
Power Density = Current Density × Cell Voltage (3.13)
While Kong et al. reported this phenomenon in cells “ignited” with unsaturated hydrocarbon additives, we initiated cell reactions using solely research-grade propane. This
approach enabled us to conduct parametric studies to investigate the deactivation dynamics
and how they are influenced by various w factors without the use of additives.
50
0 50 100 150 200 250 300 350
Time (s)
0
5
10
15
20
25
30
P
o
w
er D
e
nsity (m
W/c
m2
)
5 mA/cm2
10 mA/cm2
15 mA/cm2
20 mA/cm2
30 mA/cm2
40 mA/cm2
(a) Power dynamics
0 50 100 150 200 250 300 350
Time (s)
0.0
0.2
0.4
0.6
0.8
V
olta
g
e (V)
5 mA/cm2
10 mA/cm2
15 mA/cm2
20 mA/cm2
30 mA/cm2
40 mA/cm2
(b) Voltage dynamics
Figure 3.13: Effect of the current density on the deactivation dynamics for MEA PtB CPL
with 12.2 mg/cm2 and 9.2 mg/cm2
loadings on the anode and cathode sides, respectively.
3.4.1 Effect of current density
The deactivation dynamics for different current loads were investigated for each type of
MEA. The data was collected with the propane and oxygen flowrates set to 1.2 SLM and
0.8 SLM, respectively, and the cell temperature to 85 ◦C.
Figure 3.13 shows the results for the homemade MEA PtB CPL, and Figures 3.14 and
3.15 show the results for the commercial MEA PtC FCS and MEA PtB FCS, respectively.
For all experiments, the cell was operated with no current load until a stable OCV was
reached, then the specified current was applied at t = 0.
It is apparent that the current density significantly influences the deactivation rate of the
cell. In our tests across all MEAs, low current densities led to an expected initial drop in
voltage, which then stabilized, allowing the cell to operate continuously. At higher current
densities, the voltage continually decreased until it reached 0 V, causing the cell to stop
generating power. Initially, increasing the current density boosts the cell’s maximum power
output. However, higher current densities also lead to quicker cell extinction, where the cell
stops producing power. Consequently, lower current densities result in greater total power
output over time due to the sustained power densities. If the cell reaches extinction, it
remain in that state indefinitely, as long as the current load is maintained on it.
51
0 20 40 60 80 100 120
Time (s)
0.0
0.5
1.0
1.5
2.0
2.5
3.0
P
o
w
er D
e
nsity (m
W/c
m2
)
1 mA/cm2
3 mA/cm2
5 mA/cm2
10 mA/cm2
15 mA/cm2
Figure 3.14: Effect of the current density on the deactivation dynamics for MEA PtC FCS
with 1.5 mg/cm2
loading.
0 20 40 60 80 100
Time (s)
0.0
0.2
0.4
0.6
0.8
1.0
1.2
P
o
w
er D
e
nsity (m
W/c
m2
)
1 mA/cm2
3 mA/cm2
5 mA/cm2
10 mA/cm2
15 mA/cm2
Figure 3.15: Effect of the current density on the deactivation dynamics for MEA PtB FCS
with 4 mg/cm2
loading.
52
The best-performing MEA we tested was the homemade MEA PtB CPL, achieving instantaneous power densities of about 15 mW/cm2 and sustained power densities of 4 mW/cm2
.
Additionally, this MEA could handle high currents up to 40 mA/cm2
, with full extinction
occurring at 30 mA/cm2 within approximately 70 s, as shown in Figure 3.13.b. The deactivation rate, determined by the rate of voltage change over time, was 0.0121 V/s at 30 mA/cm2
.
The minimum voltage this cell could maintain under constant load was 0.3 V; operating
below this voltage led to extinction, a characteristic observed in other MEAs as well.
In comparison, the commercial MEAs, MEA PtC FCS and MEA PtB FCS, performed
significantly worse, with power densities much lower than those produced by MEA PtB CPL.
They also showed quicker extinction and lower sustainable current densities. Extinction
occurred at current densities greater than 1 mA/cm2
for the commercial membranes, while
MEA PtB CPL operated up to 10 mA/cm2 without extinction. We attribute this difference
to the higher anode catalyst loading in MEA PtB CPL, suggesting that catalyst deactivation
plays a critical role in these dynamics.
Our findings, shown in Figure 3.13, align well with Kong et al.’s results under similar
conditions (as seen in [59], Figure 2-b), despite Kong’s use of unsaturated hydrocarbons
to“ignite” the cell initially. This consistency between our results with pure propane and
Kong’s with additives supports the idea that Kong’s ignition challenges could stem from
testing hardware issues.
The characteristic deactivation dynamics of direct propane fuel cells make the use of
polarization curves for analysis impossible. Polarization curves, which plot voltage against
current density, are typical for assessing a PEMFC’s performance and identifying efficiency
losses, thereby guiding optimization and characterization efforts [63]. Typically, for hydrogen
PEMFC, these curves are generated by gradually increasing the current density from zero
and recording the stabilized voltage at each step. However, in direct propane PEMFC operating at sufficiently high current loads, voltage never stabilizes due to inherent deactivation
dynamics, making traditional polarization curve analysis inapplicable.
53
At high current densities, all fuel cells exhibit extinction behavior due to mass transport
losses, where the concentration of reactants near the electrodes becomes insufficient for
the reaction to continue. However, the abrupt extinction observed in our direct propane
PEMFC, occurring at unexpectedly low current densities, suggests a different cause than
mass transport limitations. Understanding this mechanism is crucial for improving direct
propane PEMFC’ viability for portable power applications by preventing extinction at high
currents and enabling operation at higher power outputs.
3.4.2 Effect of fuel flow rate
Throughout this study, we used an anode flow rate of 1.2 SLM propane and a cathode flow
rate of 0.8 SLM oxygen, following the parameters set by Kong et al. This choice was made
to facilitate a systematic comparison with their findings. However, as indicated in Figure
3.9, the flow rate of reactants might impact cell behavior. Motivated by this, we explored
how fuel flow rate influences deactivation dynamics.
Figure 3.16 presents results showing the impact of varying propane fuel flow rates on the
deactivation dynamics of a 0.5 mg/cm2 MEA PtC FCS. Across three sets of experiments, we
adjusted the propane flow rate as described in the figure legends. The key difference among
these sets of experiments was in the total anode flow rate and the cathode oxygen flow rate.
We conducted three types of experiments: one varying only the propane flow rate, another
adjusting both propane and oxygen flow rates to maintain their ratio, and a third varying
the propane flow rate while using nitrogen to keep the total anode flow rate constant at
1.2 SLM.
All three experiments showed consistent trends: increasing the propane flow rate enhanced the cell’s maximum power output and delayed extinction. This improvement plateaued
beyond a 1.2 SLM flow rate, suggesting a threshold for fuel flow rate benefits. The consistent
trend across all three experiments suggests that the flow rate changes deactivation dynamics
chemically rather than mechanically, as illustrated in Figure 3.16.c. Here, even when the
54
0 2 4 6 8 10 12
Time (s)
0.0
0.2
0.4
0.6
0.8
1.0
P
o
w
er D
e
nsity (m
W/c
m2
)
0.1 SLM
0.5 SLM
0.8 SLM
1.2 SLM
1.5 SLM
2.0 SLM
(a) Varying propane flow rate
0 2 4 6 8 10 12 14 16
Time (s)
0.0
0.2
0.4
0.6
0.8
1.0
P
o
w
er D
e
nsity (m
W/c
m2
)
0.1 SLM
0.5 SLM
0.8 SLM
1.2 SLM
1.5 SLM
2.0 SLM
(b) Varying propane and oxygen flow rates
0 2 4 6 8 10 12 14 16
Time (s)
0.0
0.2
0.4
0.6
0.8
1.0
P
o
w
er D
e
nsity (m
W/c
m2
)
0.1 SLM
0.5 SLM
0.8 SLM
1.2 SLM
(c) Varying propane flow rate, keeping anode flow
rate constant
Figure 3.16: The effect of propane fuel flow rate at the anode on the deactivation dynamics of
a 0.5 mg/cm2 MEA PtC FCS under a constant load of 3 mA/cm2
. (a) Anode flow consisting
solely of propane, varied as indicated in the legend, with a constant cathode flow rate of
0.8 SLM oxygen. (b) Anode flow consisting solely of propane, adjusted as per the legend,
with oxygen cathode flow rate also varied to maintain an anode:cathode flow rate ratio of
1.5. (c) Anode flow composed of propane and nitrogen; propane flow rate varied as shown
in the legend, with nitrogen supplementing the anode flow to maintain a total flow rate of
1.2 SLM, keeping the cathode flow rate constant at 0.8 SLM oxygen.
55
propane concentration in the anode flow rate was varied with nitrogen to maintain a constant
flow, cell performance still improved, indicating that the effect is not due to changes in shear
stress or water content in the cell. This behavior differs from that observed with hydrogen
fuel, where the cell’s OCV and maximum power density were unaffected by fuel flow rate
as long as sufficient fuel was supplied. Relative to the power densities of the level, propane
flow rate of 0.1 SLM should theoretically be sufficient.
All the 3 sets of experiments show the same trends. We observed that increasing the
propane fuel flow rate resulted in increasing the maximum power generated by the cell and
delayed time until cell extinction. This effect persisted up to a point, and increasing the
propane flow rate further beyond 1.2 SLM, did not result in further improvement. The fact
that trend is consistent across all 3 sets of experiments indicates that increasing the flow
rates changes the deactivation dynamics through chemical means rather than mechanical
means. This is evident by Figure 3.16.c where we vary the the concentration of propane
in the anode flow rate while keeping the anode flow rate constant by using supplemental
nitrogen. We still observe improvement in cell performance. We cannot explain the effect by
the change in the shear stress or the total water content carried into the cell. Such behaviour
was not observed when we operated the same cell with hydrogen as the fuel. In the case
of hydrogen, the cell OCV was independent of the fuel flowrate, and the maximum power
density of the cell was not affected by the fuel flowrate as long as enough fuel was supplied
to generate the required power. Relative to the power densities of the level, propane flow
rate of 0.1 SLM should theoretically be sufficient.
All three sets of experiments revealed consistent trends. Increasing the propane fuel flow
rate enhanced the cell’s maximum power output and postponed the onset of cell extinction.
This improvement plateaued when the propane flow rate exceeded 1.2 SLM, showing no
further gains. The consistent pattern across all experiments suggests that the flow rate
affects deactivation dynamics chemically rather than mechanically. This is supported by
Figure 3.16.c, where varying the concentration of propane while maintaining the anode flow
56
rate constant with supplemental nitrogen still led to performance enhancements. These
effects cannot be attributed to changes in shear stress or water content within the cell. When
operating the cell with hydrogen as the fuel, such behavior was absent. With hydrogen, the
cell’s OCV remained unaffected by the fuel flow rate, and the cell’s maximum power density
was stable as long as the fuel supply was sufficient relative to the current load. Theoretically,
a propane flow rate as low as 0.1 SLM should be sufficient for the current loads we applied
on the cell.
The results from Figures 3.16.a and 3.16.b, indicate that operating the cell at 1.2 SLM
propane is sufficient for optimal performance.
3.4.3 Effect of catalyst loading
The performance of direct propane PEMFC is notably limited by propane’s reactivity in
the cell, leading us to hypothesize that increased catalyst loading could mitigate or slow
down cell deactivation. Despite the cost limitations associated with higher catalyst loadings,
we generally opted for higher loadings in our research to investigate the maximum performance potential of direct propane PEMFC. However, we conducted the following study to
investigate the effect of varying the loading on the deactivation dynamics.
Figure 3.17 illustrates a comparison of direct propane deactivation dynamics as PtC loading is increased from 0.5 to 1.5 mg/cm2
in a MEA PtC FCS. Consistent with expectations,
cell performance saw improvement. Both the initial instantaneous power output and the
duration until extinction approximately doubled for specific current density values. This
underscores that enhancing the loading improves performance, but the impact is not linear
and may reach a saturation point beyond a certain loading threshold. Kong et al. reported
similar trends with their cell [58].
Furthermore, examining the performance difference between the 4 mg/cm2 MEA PtB FCS
and the 12.2 mg/cm2 MEA PtB CPL, as shown in Figures 3.15 and 3.13, is insightful. Although the two MEA variants differ in several aspects beyond catalyst loading, it’s likley
57
0 20 40 60 80 100 120 140
Time (s)
0.0
0.5
1.0
1.5
P
o
w
er D
e
nsity (m
W/c
m2
)
1 mA/cm2
3 mA/cm2
5 mA/cm2
10 mA/cm2
15 mA/cm2
(a) 0.5 mg/cm2
0 20 40 60 80 100 120
Time (s)
0.0
0.5
1.0
1.5
2.0
2.5
3.0
P
o
w
er D
e
nsity (m
W/c
m2
)
1 mA/cm2
3 mA/cm2
5 mA/cm2
10 mA/cm2
15 mA/cm2
(b) 1.5 mg/cm2
Figure 3.17: The effect of catalyst loading on the deactivation dynamics for MEA PtC FCS
for two different loading’s: (a) 0.5 mg/cm2 and (b) 1.5 mg/cm2
.
that the better performance of MEA PtB CPL is primarily attributed to its high anode catalyst loading. Here, ’better performance’ denotes higher power densities for specific current
loads and a delayed extinction. In this context, tripling the catalyst loading has led to an
improvement exceeding threefold, particularly when considering both the instantaneous and
sustained power generated at various current loads. It is important to notice, however, that
supported platinum catalyst achieves more catalyst utilization than platinum black [64].
3.4.4 Effect of cell temperature
We chose the nominal operating cell temperature to be 85 ◦C because such temperature is
suitable for portable power applications such as in consumer electronics.Additionally, our
cell exhibited its highest open-circuit voltage OCV at 85 ◦C, as shown in Figure 3.12. The
variation of OCV with cell temperature suggests a similar dependence of deactivation dynamics on temperature. Motivated by this observation, we explored the impact of different
temperatures on deactivation dynamics.
The cell temperature is controlled by two adhesive heating pads attached to the outside
of the current collector plates. A thermocouple that penetrates the graphite plates measures
the cell temperature and provides feedback for the heater controller. Figure 3.18 shows the
effect of varying the direct propane PEMFC cell temperature on the deactivation dynamics.
58
10
0 10
1 10
2 10
3
Time (s)
0.00
0.25
0.50
0.75
1.00
1.25
1.50
P
o
w
er D
e
nsity (m
W/c
m2
)
65°C
75°C
85°C
95°C
(a) MEA PtC FCS
10
0 10
1 10
2 10
3
Time (s)
0
5
10
15
20
P
o
w
er D
e
nsity (m
W/c
m2
)
55 °C
70 °C
85 °C
95 °C
104 °C
(b) MEA PtB CPL
Figure 3.18: The effect of cell temperature on the deactivation dynamics for 0.5 mg/cm2
MEA PtC FCS and 12.2 mg/cm2 MEA PtB CPL under a constant current load of 3 mA/cm2
and 30 mA/cm2
for (a) and (b), respectively. Note the logarithmic timescale.
Increasing the cell temperature improved cell performance by increasing the max power
density obtained from the cell and extending the time duration until extinction. This effect is
saturated after 85 ◦C for the MEA PtC FCS, and increasing the temperature further seems
to have no significant effect. For the MEA PtB CPL, performance continued to improve
as the temperature was increased up to 104 ◦C. Higher temperatures were not tested due
to limitations in the heating pads’ power. Kong et. al. observed similar trends in their
propane PEMFC using platinum black, noting that increasing the temperature up to 114 ◦C
enhanced maximum power density and overall performance [58].
The improvement in performance at higher temperatures was particularly notable for
the MEA PtB CPL. At 104 ◦C, the cell operated at 30 mA/cm2
for 1000 s before reaching extinction, producing an average power of 8 mW/cm2
. These results encouraged us to
further investigate the deactivation dynamics of direct propane PEMFC operating at high
temperatures, as detailed in the following section.
In hydrogen PEMFC, cell temperature also positively correlates with performance [65].
It has been concluded that increasing the cell’s temperature enhances membrane conductivity and accelerates the electrochemical reaction rate, thereby improving performance. We
suspect that similar factors influence the performance of propane PEMFC under varying
59
10
0 10
1 10
2 10
3
Time (s)
0
10
20
30
40
50
P
o
w
er D
e
nsity (m
W/c
m2
)
30 mA/cm2
40 mA/cm2
50 mA/cm2
60 mA/cm2
70 mA/cm2
80 mA/cm2
Figure 3.19: Effect of the current density on the deactivation dynamics for MEA PtB CPL
with 12.2 mg/cm2 anode catalyst loading and operating at 104 ◦C. Note the logarithmic
timescale.
cell temperatures, particularly the reaction rate, which we hypothesize to be the limiting
performance factor for propane PEMFC.
3.4.5 Deactivation dynamics at high cell temperatures
Motivated by the results from Figure 3.18.b, we investigated the dynamics of direct propane
PEMFC operating at 104 ◦C. It is important to highlight that at temperatures above 100 ◦C,
the PEMFC environment lacks liquid water, leaving only gaseous water. This condition is
expected to significantly impact the PEMFC reaction kinetics and deactivation dynamics.
Figure 3.19 displays the deactivation dynamics at various constant current loads for a
MEA PtB CPL cell operating at 104 ◦C. These results can be contrasted with those in
Figure 3.13, which shows the same cell at 85 ◦C. Raising the cell temperature to 104 ◦C
increases both the instantaneous and average power output for specific current densities and
extends the time until extinction. Moreover, at this elevated temperature, we successfully
operated the cell at significantly higher current densities, up to 80 mA/cm2
, which were not
achievable at 85 ◦C.
These results confirm that operating the cell at a higher temperature is desirable. How60
ever, for portable power generation applications, such high temperature operation may pose
practical challenges.
3.4.6 Summary of MEAs performance
In this section, we explored the deactivation dynamics of three distinct MEA types: MEA PtC FCS,
MEA PtB FCS, and MEA PtB CPL. All three demonstrated similar general deactivation
trends and were influenced similarly by various operating conditions.
The commercial MEA PtC FCS, the most cost-effective option among those we examined, closely resembles standard MEAs found in typical hydrogen PEMFC applications.
Despite catalyst loadings as high as 1.5 mg/cm2
, the MEA PtC FCS achieved power outputs around 2 mW/cm2
, which is roughly 100 times lower than the power generated by
comparable cells fueled by hydrogen.
The commercial MEA PtB FCS, with a catalyst loading of 4 mg/cm2
, exhibited performance comparable to that of the MEA PtC FCS. This similarity in performance may stem
from the enhanced catalyst utilization provided by the supported platinum black. Both
MEA types exhibited similar deactivation dynamics.
The homemade MEA PtB CPL outperformed the commercial membranes significantly,
producing more power and sustaining power output for longer periods, even under higher
current loads. We hypothesize that the enhanced performance primarily results from the
higher anode catalyst loading of 12.2 mg/cm2
. Based on these results, the MEA PtB CPL
configuration appears to be the most suitable for direct propane PEMFC. Despite the associated increase in fuel cell cost attributed to the higher catalyst loading, the benefits are
anticipated to outweigh the drawbacks for portable power generation applications. This is
because portable application require less active cell area, as the total power requirement is
relatively small, thereby offsetting the cost of the catalyst.
61
3.4.7 Conclusions
Despite advancements in characterizing deactivation dynamics of direct propane PEMFC
and our understanding of them, the fundamental mechanisms remain unclear. Operational
factors such as current density, fuel flow rate, catalyst loading, and temperature influence
deactivation and cell extinction, yet do not fully explain the observed abrupt voltage declines. This indicates a complex deactivation mechanism beyond traditional mass transport
limitations, potentially involving catalyst poisoning, membrane degradation, or specific interactions between propane and other intermediates present on or near the reaction sites.
The performance difference among different MEAs, especially the significantly better performance of the homemade MEA PtB CPL due to higher anode catalyst loading, suggests a
significant role of catalyst-related phenomena in deactivation. However, the exact contributions of these phenomena to the rapid voltage and power decline are not fully understood.
To advance the application of direct propane PEMFC in portable power, understanding
these deactivation mechanisms at a molecular or cell level is crucial. Future research should
focus on identifying the root causes of deactivation through advanced characterization and
theoretical modeling. Understanding the root causes of deactivation will allow us to design
more robust and efficient fuel cells capable of sustained operation under diverse conditions.
3.5 Activation dynamics
In the preceding discussion, we observed a phenomenon in the operation of the direct propane
PEMFC under constant and sufficiently high current loads: cell deactivation, characterized
by a voltage decline to 0 V. Upon reduction or cessation of the current load, the cell voltage
initiates recovery. Notably, in the absence of any current load, the cell’s OCV is restored
to its original value. This phenomenon, we term “activation,” describes the activation of
the cell’s reaction sites subsequent to the cell’s full or partial deactivation, induced by the
reduction or elimination of the current load. Since we are unable to directly measure fraction
62
of active reaction sites in the cell, we use the cell’s voltage dynamics as proxy measure of
activation dynamics.
Kong et al.’s previous work noted that a direct propane fuel cell could be “reset” by
removing the current load after reaching extinction (0 V), thereby recovering the original
OCV. After resetting, reapplying the current load allowed them to observe the deactivation
process until the cell reached extinction again. They proposed exploiting this phenomenon
in a cell operation mode termed the “load-interrupt” mode (Figure 3.3), where a current
load is applied for 20 s and then removed for 5 s in a repeated cycle.
Motivated by the observation of this activation phenomena and its potential to shed light
on the deactivation mechanism taking place and its use as a possible operation mode, we
investigated the activation dynamics. In this section, we describe our investigation on the
activation dynamics and how different operation factors affect them. We limit our discussion
to the MEA PtB CPL fabricated in-house, since that MEA had the most potential for direct
propane PEMFC.
3.5.1 Effect of current density
The current density has a strong influence on deactivation dynamics. This motivated us
to explore how various current loads, leading to full or partial cell deactivation, impact the
subsequent activation dynamics once the current is removed.
We conducted a series of experiments on the MEA PtB CPL, as shown in Figure 3.20,
to understand the effect of current density on activation dynamics. For each experiment,
we first applied a constant current load on the cell, as specified in the figure legend. As
a result, the cell deactivated, and the voltage dropped. Once the voltage dropped below
0.1 V or reached a stable value, the current load was then removed (0 mA/cm2
). With the
current removed, we observed the voltage rise over time, indicating the activation dynamics
as shown in the figure. It’s noteworthy that for current values of 5 mA/cm2 and 10 mA/cm2
,
extinction did not occur, and the cell reached a stable voltage value in the range of 0.3 V to
63
10
0 10
1 10
2 10
3
Time (s)
0.0
0.2
0.4
0.6
0.8
V
olta
g
e (V)
5 mA/cm2
10 mA/cm2
15 mA/cm2
30 mA/cm2
40 mA/cm2
Figure 3.20: Effect of the current density during deactivation on the activation dynamics for
MEA PtB CPL with 12.2 mg/cm2 anode catalyst loading and operating at 85 ◦C. Note the
logarithmic timescale.
0.45 V. For these two cases, the current load was removed once the cell voltage stabilized.
For all tested current density values, the cell recovered 75% (0.6 V) of its peak OCV
value in less than 14 s. Comparing Figure 3.20 with Figure 3.13, it is evident that activation dynamics are approximately an order of magnitude faster than deactivation dynamics.
Moreover, while deactivation is strongly dependent on current density, with current density
values between 15 mA/cm2
to 40 mA/cm2 altering the time to extinction from 40 s to 250 s,
activation dynamics exhibit a weaker dependence on current.
To investigate the current effect, we can define an activation rate and an activation
duration. The activation rate, shown in Figure 3.21.a, is the average rate of change of
voltage between the lowest voltage value and 0.6 V, which represents 75% of the cell’s stable
OCV (0.8 V). This rate should provide a normalized measure of how fast the cell recovers
its activated state. The activation duration, shown in Figure 3.21.b, is the time it took for
the cell’s voltage to increase from the smallest observed value to 0.6 V. This measure is
necessary since the lowest voltage value for each current density differs, especially for low
current densities where full deactivation does not take place.
Looking at Figure 3.21, we can observe that the cell recovers its OCV value at an average
of approximately 0.07 V/s. This rate depends on the current density load before activation
64
5 10 15 20 25 30 35 40
Current Density mA/cm2
0.04
0.06
0.08
0.10
Activation Rate (V/s)
(a) Activation rate
5 10 15 20 25 30 35 40
Current Density mA/cm2
4
6
8
10
12
Time until voltage recovery (s)
(b) Activation duration
Figure 3.21: Effect of the current density during deactivation on the activation rate and activation duration for MEA PtB CPL with 12.2 mg/cm2 anode catalyst loading and operating
at 85 ◦C. Both the activation rate and the activation duration were calculated between the
lowest observed voltage value and 0.6 V.
starts. For higher current density loads, the activation rate increases. Additionally, the
activation duration depends on the current density load before activation starts. For current
densities below 15 mA/cm2
, the cell was reactivated in less than 5 s, mainly due to the
fact that voltage did not significantly change for those low current density values. For
current densities above 15 mA/cm2
, the activation duration decreased with increasing current
density. This suggests that cells reactivate faster if the current density during deactivation
was higher. It is important to note, however, that at higher current densities, the cell operates
for shorter periods of time before reaching full deactivation. This affects the total power and
the total charge produced by the cell. We will investigate these effects in subsequent sections.
3.5.2 Effect of cell voltage before reactivation
Figure 3.20 shows that different current densities lead to varying low voltage values and subsequent differences in activation dynamics. Our objective was to systematically investigate
the influence of the cell’s voltage before activation on these dynamics, while isolating the
effect of the current. Specifically, we aimed to determine if there are notable changes in cell
dynamics between full deactivation (< 0.1 V) and partial deactivation (0.1 V to 0.5 V).
65
10
0 10
1 10
2 10
3
Time (s)
0.2
0.4
0.6
0.8
V
olta
g
e (V)
0.05 V
0.1 V
0.2 V
0.3 V
0.4 V
Figure 3.22: Effect of the cell’s voltage before activation is initiated on the activation dynamics for MEA PtB CPL with 12.2 mg/cm2 anode catalyst loading and operating at 85 ◦C.
Note the logarithmic timescale.
To conduct the investigation, a series of experiments were performed where the cell
was deactivated under a constant current load of 30 mA/cm2
. During this load, the cell’s
voltage dropped. Upon reaching a specific predetermined voltage value, the current load was
removed, and the activation dynamics were observed. The results are shown in Figure 3.22,
with the voltage values at which activation was initiated shown in the figure legend.
Similar to the previous section, we can calculate the activation rate and the activation
duration from the specific voltage at the beginning of activation to the point when the cell
voltage reaches 0.6 V. Figure 3.23 displays these results. It is observed that for the same
current load, the activation rate decreases with an increase in the voltage at the beginning
of activation. The activation duration also decreases as the pre-activation voltage increases,
because the cell has a smaller voltage range to recover. However, the difference in activation
duration ranges from 1.1 s to 2.1 s, while the activation rate is between 0.15 V/s to 0.4 V/s.
These observations confirm previous findings indicating that driving the cell to full deactivation, either through applying higher current loads or by allowing the loads to persist until
the voltage drops further, speeds up the activation dynamics.
It is important to highlight the differences in activation dynamics observed in Figure 3.22
compared to those in Figure 3.20 at 30 mA/cm2
. For the results in Figure 3.22, activation
66
0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40
Voltage before reactivation (V)
0.15
0.20
0.25
0.30
0.35
0.40
Reactivation Rate (V/s)
(a) Activation rate
0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40
Voltage before reactivation (V)
1.2
1.4
1.6
1.8
2.0
Time until voltage recovery (s)
(b) Activation duration
Figure 3.23: Effect of the cell’s voltage before activation is initiated on the activation rate
and activation duration for MEA PtB CPL with 12.2 mg/cm2 anode catalyst loading and
operating at 85 ◦C. Both the activation rate and the activation duration were calculated
between the voltage value specified on the x-axis and 0.6 V.
was initiated automatically by removing the current once the voltage reached a specified
value. In contrast, the activation in Figure 3.20 was initiated manually once the researcher
determined the cell was fully deactivated or had achieved a stable voltage. Consequently,
activation in the manual case typically did not start as fast as in the automated case. The
variance between these two sets of experiments may be due to the duration the cell remained
in a fully deactivated state before the current load was removed and activation initiated. This
aspect is explored in the subsequent section.
3.5.3 Effect of time in full deactivation state
We conducted a controlled set of experiments to investigate whether the time duration the
cell spends in a fully deactivated state affects the activation dynamics. A constant current
load of 30 mA/cm2 was applied to the cell until it was fully deactivated (V < 0.1). The
current was then maintained on the cell for a specific duration before being removed, and
the cell’s activation dynamics were observed. The activation results from these experiments
are shown in Figure 3.24. Additionally, the activation rate and activation duration for the
same experiments are presented in Figure 3.25.
67
10
0 10
1 10
2 10
3
Time (s)
0.0
0.2
0.4
0.6
0.8
V
olta
g
e (V)
0 s
3 s
10 s
120 s
Figure 3.24: Effect of the duration the cell spends in deactivated state on the activation
dynamics for MEA PtB CPL with 12.2 mg/cm2 anode catalyst loading and operating at
85 ◦C. Note the logarithmic timescale.
It was observed that leaving the cell in a fully deactivated state for 10 s or less did not
significantly affect the activation dynamics. The cell recovered an OCV of 0.6 V in less than
2 s. However, leaving the cell in a deactivated state for 120 s increased the time to recover
0.6 V to 11 s. This is evident in Figure 3.24, where the voltage for the 120 s experiments
shows a slower recovery rate at the beginning of the activation process. The activation rate
decreased for longer deactivated periods. This indicates that the cell’s reaction sites undergo
some changes while the cell remains deactivated, and the voltage is relatively stable below
0.1 V. These results suggest that voltage might not be a perfect measure for the activation
state of the cell, as expected.
Additionally, these results clarify the difference observed earlier between Figure 3.22 and
Figure 3.20 at 30 mA/cm2
. A longer period before initiating activation by removing the
current load slows down the activation dynamics. The findings imply that once the cell
is deactivated and ceases to produce power, it might be beneficial to reactivate the cell
immediately to regain the OCV faster and exploit the quicker activation dynamics.
68
0 20 40 60 80 100 120
Time at extinction (s)
0.1
0.2
0.3
0.4
Reactivation Rate (V/s)
(a) Activation rate
0 20 40 60 80 100 120
Time at extinction (s)
2
4
6
8
10
Time until voltage recovery (s)
(b) Activation duration
Figure 3.25: Effect of the duration the cell spends in deactivated state on on the activation
rate and activation duration for MEA PtB CPL with 12.2 mg/cm2 anode catalyst loading
and operating at 85 ◦C. Both the activation rate and the activation duration were calculated
between the lowest voltage value observed and and 0.6 V.
3.5.4 Effect of total energy and total charge
The results in Figure 3.20 demonstrated that the current load before activation impacts
activation dynamics. However, the current load also affects the time until extinction, thereby
altering the total power and total charge produced by the cell. Motivated by this, we
conducted a controlled set of experiments to investigate the effects of total energy and total
charge on the activation dynamics.
We operated the cell under a low constant current load of 10 mA/cm2
, which does not
lead to full deactivation or cell extinction. The load was applied for a specific duration before
being removed, and then the cell’s activation was observed. The results of the activation
dynamics from this experiment are shown in Figure 3.26. The activation duration as a
function of total power and total charge was also calculated and is presented in Figure 3.27.
As hypothesized in previous sections, the activation dynamics seem to strongly depend
on the total power and charge produced by the cell. Operating the cell for longer periods
under a constant current load that does not cause cell extinction increased the total power
generated by the cell, and as a result, the time it took for the cell to recover a voltage
of 0.6 V increased. The relationship between the total power produced and the activation
69
0 100 200 300 400 500
Time (s)
0
2
4
6
P
o
w
er D
e
nsity (m
W/c
m2
)
60 s
120 s
480 s
(a) Power dynamics during deactivation
10
0 10
1 10
2 10
3
Time (s)
0.3
0.4
0.5
0.6
0.7
0.8
V
olta
g
e (V)
60 s
120 s
480 s
(b) Activation dynamics
Figure 3.26: Effect of the total power produced by the cell on the activation dynamics for
MEA PtB CPL with 12.2 mg/cm2 anode catalyst loading and operating at 85 ◦C. In (a), the
cell was operated under a constant current load of 10 mA/cm2
. Note the logarithmic time
scale in (b).
duration is linear. Similarly, a similar relationship is observed between the total charge
and the activation duration. This is expected since the total power and total charge are
strongly correlated, given that the voltage at a current value of 10 mA/cm2 does not decrease
significantly during the tested durations.
These results help explain those shown in Figure 3.20, where higher current loads resulted in quicker activation dynamics. The faster dynamics are likely due to higher current
loads leading to quicker extinction and consequently less total power production by the cell.
Therefore, operating the cell for extended periods at lower current loads will slow down
activation dynamics, which is not desirable.
3.5.5 Effect of cell temperature
Similar to our investigation of the cell’s temperature effect on deactivation dynamics, we
explored whether temperature influences activation dynamics. We operated the cell at a
constant load of 30 mA/cm2 until full deactivation occurred, indicated by the cell voltage
dropping below 0.1 V. Following this, we removed the current to observe the activation
dynamics. This procedure was repeated across different cell temperatures, with the results
70
200 400 600 800 1000 1200 1400
Total Energy Produced (mJ/cm2
)
2.0
2.2
2.4
2.6
2.8
3.0
3.2
Time until voltage recovery (s)
(a) Activation duration
1000 2000 3000 4000 5000
Total Charge (mC/cm2
)
2.0
2.2
2.4
2.6
2.8
3.0
3.2
Time until voltage recovery (s)
(b) Activation duration
Figure 3.27: Effect of the total power and total charge produced by the cell on the activation
duration for MEA PtB CPL with 12.2 mg/cm2 anode catalyst loading and operating at
85 ◦C. The activation duration were calculated between the lowest voltage value observed
and 0.6 V.
shown in Figure 3.28. The activation rate and activation duration for this set of experiments
were calculated, and the results are displayed in Figure 3.29.
Similar to the impact of temperature on deactivation dynamics, an increase in cell temperature accelerated the activation dynamics. Operating the cell at higher temperatures
improved the activation rate and reduced the time needed for the cell to recover to 0.6 V.
Operating the cell at 104 ◦C resulted in slower activation dynamics than at 95 ◦C, possibly
due to the absence of liquid water affecting chemical kinetics. It is noteworthy that there
was a linear correlation between cell temperature and total power output, as higher temperatures allowed the cell to operate longer before reaching full deactivation. Yet, the increase
in temperature also hastened the activation dynamics, even as total power increased. Compared to the findings in Figure 3.26, this suggests that the temperature’s effect on activation
dynamics outweighs that of total power generated before activation begins.
While the higher temperature likely slowed deactivation dynamics by enhancing the reaction rates of the propane reaction, the reason for accelerated activation dynamics at higher
temperatures remains uncertain. One hypothesis is that higher temperatures expedite the
reactivation of catalyst sites by increasing product desorption rates at reaction sites. These
results further suggest that operating the cell at higher temperatures is preferable from a
71
10
0 10
1 10
2 10
3
Time (s)
0.0
0.2
0.4
0.6
0.8
V
olta
g
e (V)
55 °C
70 °C
85 °C
95 °C
104 °C
Figure 3.28: Effect the cell temperature on the activation dynamics for MEA PtB CPL with
12.2 mg/cm2 anode catalyst loading. The cell was deactivated using a constant current load
of 30 mA/cm2
. Note the logarithmic timescale.
performance standpoint, given that quicker activation dynamics are advantageous.
3.5.6 Conclusions
Among all operating factors considered here, it is apparent that activation dynamics are
significantly faster than deactivation dynamics. This suggests that the processes involved in
the activation and deactivation of reaction sites operate on different time scales. We observed
that activation was strongly influenced by the total power and charge generated by the cell,
with more power and charge slowing the activation dynamics. Additionally, operating the
cell at higher temperatures accelerated the activation dynamics.
Based on the results presented in this section, to increase the activation rate after full
deactivation, it is advisable to reach full deactivation quickly and at relatively higher current densities. Once full deactivation is achieved, the activation process should be initiated
immediately. The activation dynamics need to be considered when devising any cyclic operation strategy for direct propane PEMFC, as fast activation dynamics are highly desirable
to increase the cell’s average power density.
72
60 70 80 90 100
Fuel Cell Temperature (C)
0.1
0.2
0.3
0.4
Reactivation Rate (V/s)
(a) Activation rate
60 70 80 90 100
Fuel Cell Temperature (C)
2
4
6
8
10
Time until voltage recovery (s)
(b) Activation duration
Figure 3.29: Effect of the cell temperature on the activation rate and activation duration for
MEA PtB CPL with 12.2 mg/cm2 anode catalyst loading. Both the activation rate and the
activation duration were calculated between the lowest voltage value observed and 0.6 V.
3.6 Effect of fuel additives on cell performance
Kong et al. observed that adding small amounts of chemical additives to the fuel stream
delayed or prevented the extinction of the direct propane PEMFC and increased the power
density [59]. They specifically found that unsaturated hydrocarbons, added in quantities
in the range of 470 PPM to 2540 PPM, such as ethylene, had this effect. Motivated by
these findings, we systematically investigated the impact of adding ethylene and hydrogen
to the fuel stream of our PEMFC running on research-grade propane. In this section, a fuel
additive refers specifically to substances added to the anode fuel stream, which consists of
research-grade propane, to alter the dynamics of the direct propane PEMFC.
3.6.1 Effect of ethylene as a fuel additive
In [59], researchers determined that ethylene, among various unsaturated hydrocarbons,
was the most effective additive for enhancing the power density of direct propane PEMFC
and mitigating extinction. This prompted further investigation into the impact of varying
ethylene concentrations on our PEMFC dynamics fueled by research-grade propane.
73
0 50 100 150 200 250 300 350 400
Time (s)
0.0
0.5
1.0
1.5
2.0
2.5
3.0
Ethylene flowrate (SCCM)
(a) Ethylene flowrate
0 50 100 150 200 250 300 350 400
Time (s)
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
V
olta
g
e (V)
(b) Cell OCV
Figure 3.30: Effect of adding 3 SCCM of ethylene (2500 PPM) to the anode fuel stream of
the MEA PtB CPL on the cell’s OCV. The anode and cathode flow streams were 1.2 SLM
propane and 0.8 SLM oxygen, respectively. (a) The ethylene flow rate over time; we removed
the ethylene flow rate temporarily around t = 300 s. (b) The cell’s OCV.
3.6.1.1 Effect on OCV
As discussed in section 3.3 (Figure 3.11), the MEA PtB CPL had a maximum OCV of 0.9 V
and a stable OCV of approximately 0.7 V that is reached after more than 1000 s when the
cell is fueled by propane alone. Using the same MEA, we investigated the effect of adding
ethylene to the propane fuel stream on the cell’s OCV, and the results are shown in Figure
3.30.
We operated the cell under no current load and observed the OCV. The anode flow rate
was started at t = 0 and consisted of 1.2 SLM propane and 3 SCCM ethylene (2500 PPM).
As the cell initiated, the OCV climbed steadily and reached a peak value of approximately
0.8 V before decreasing to 0.7 V. At approximately t = 300 s, we shut off the ethylene flow
and kept the propane flow. It was observed that the OCV instantly increased and eventually
reached a steady value of around 0.8 V. At approximately t = 375 s, we resumed the ethylene
flow rate and the OCV decreased to 0.7 V again. The ethylene flow rate is shown in Figure
3.30.a. Different OCV experiments not shown here indicate that a direct propane fuel cell
with 3 SCCM ethylene reaches a stable OCV of 0.7 V to 0.8 V.
The findings show that ethylene addition does not elevate OCV, implying no improvement
74
in cell’s reaction potential from ethylene. Initially, ethylene reduces OCV, but this effect
dissipates over time.
Comparison with [59] is limited as they did not report long-term OCV behavior or improvement without additives. Their observed OCV of 0.8 V with ethylene aligns with our
findings [59]-fig 2.a.
3.6.1.2 Effect on cell performance
After investigating the OCV, we systematically investigated the effect of adding various
quantities of ethylene to the fuel stream on the cell’s performance. Specifically, we studied
how the deactivation dynamics of the cell are affected by ethylene. Previous work showed
promise that low quantities of ethylene can increase the cell’s power output and prevent full
deactivation.
Testing on the MEA PtB CPL, we added ethylene in quantities of 2500, 8300, and
14 700 PPM to a constant anode flow rate of 1.2 SLM propane, and we observed the deactivation dynamics under different constant loads. The results for those experiments are
shown in Figure 3.31.
Compared to the results shown in 3.13 for the deactivation dynamics of MEA PtB CPL
operating only on propane, the propane-ethylene mixture resulted in different deactivation
dynamics. For low current values (less than 10 mA/cm2
), adding ethylene to the propane
did not significantly change the deactivation dynamics. However, at higher current densities
(more than 20 mA/cm2
), the addition of ethylene decreased the initial power produced by
the cell but prevented full deactivation or cell extinction. We did not observe extinction
for any of the current values reported, up to 60 mA/cm2
. As a result of preventing full
deactivation, we were able to operate the cell at a steady state condition at a higher current
density load and generate more power, compared to using only propane. The steady-state
power densities were linearly correlated with the amount of ethylene. Namely, adding more
ethylene allowed us to operate under higher current loads and generate more power as a
75
result. This effect saturated after a certain amount of ethylene, and in general, the highest
steady-state power densities we were able to observe were 7.5 mW/cm2
.
Because the cell operating on propane and ethylene did not exhibit full deactivation,
we were able to perform standard polarization curve tests, and the results are shown in
Figure 3.33. The polarization curve shows that power densities up to 9 mW/cm2 and current
densities up to 80 mA/cm2 are possible in the cell. We found that such values are not
attainable when operating the cell with research-grade propane alone.
To investigate whether the power production in the propane-ethylene cells was due to
both fuels or the ethylene, we conducted a similar set of experiments to those in Figure
3.31, but for ethylene in nitrogen. Namely, we added ethylene in quantities of 2500, 8300,
and 14 800 PPM to a constant anode flow rate of 1.2 SLM nitrogen, and we observed the
deactivation dynamics under different constant loads. We replaced the propane with the
same amount of nitrogen to keep the mechanical environment inside the cell constant, and
since nitrogen is an inert gas, we can be certain that it is not contributing to any power
produced by the cell. As a result, any power production will be generated by the ethylene
oxidation reaction only. The results for those experiments are shown in Figure 3.32.
Comparing the results in Figures 3.31 and 3.32, it becomes apparent that cell performance, especially after stabilization, is highly similar whether the cell operates on a mixture of propane and ethylene or ethylene and nitrogen. This observation suggests that,
post-stabilization, power production is predominantly due to ethylene in the fuel stream,
regardless of propane’s presence. It’s possible that while the propane reaction continues
to deactivate independently, the ethylene reaction remains active and continues generating
power. Notably, there are differences between the results in the initial operation period of
the cell. For some current values, the propane/ethylene cell exhibits higher power densities, likely due to propane initially reacting and contributing to power production before its
reaction fully deactivates.
Kong et al. argued in [59] that their cell operating on a propane-ethylene mixture must
76
0 100 200 300 400 500
Time (s)
0.0
2.5
5.0
7.5
10.0
12.5
15.0
P
o
w
er D
e
nsity (m
W/c
m2
)
10 mA/cm2
20 mA/cm2
30 mA/cm2
(a) 3 SCCM 2500 PPM
0 20 40 60 80 100 120
Time (s)
0.0
2.5
5.0
7.5
10.0
12.5
15.0
P
o
w
er D
e
nsity (m
W/c
m2
)
10 mA/cm2
20 mA/cm2
30 mA/cm2
40 mA/cm2
(b) 10 SCCM 8300 PPM
0 25 50 75 100 125 150 175
Time (s)
0.0
2.5
5.0
7.5
10.0
12.5
15.0
P
o
w
er D
e
nsity (m
W/c
m2
)
10 mA/cm2
20 mA/cm2
30 mA/cm2
40 mA/cm2
60 mA/cm2
(c) 18 SCCM 14 800 PPM
Figure 3.31: Ethylene in Propane.
The effect of adding different quantities
of ethylene, as indicated in the subfigures’ labels, to the anode flow rate of
1.2 SLM propane on the deactivation dynamics for MEA PtB CPL under constant
current load. The current loads are specified
in the sub-figures’ legends. The cathode flow
rate was 0.8 SLM oxygen.
0 50 100 150 200
Time (s)
0.0
2.5
5.0
7.5
10.0
12.5
15.0
P
o
w
er D
e
nsity (m
W/c
m2
)
10 mA/cm2
20 mA/cm2
30 mA/cm2
40 mA/cm2
(a) 3 SCCM 2500 PPM
0 50 100 150 200 250 300
Time (s)
0.0
2.5
5.0
7.5
10.0
12.5
15.0
P
o
w
er D
e
nsity (m
W/c
m2
)
10 mA/cm2
20 mA/cm2
30 mA/cm2
40 mA/cm2
(b) 10 SCCM 8300 PPM
0 100 200 300 400 500 600
Time (s)
0.0
2.5
5.0
7.5
10.0
12.5
15.0
P
o
w
er D
e
nsity (m
W/c
m2
)
10 mA/cm2
20 mA/cm2
30 mA/cm2
40 mA/cm2
60 mA/cm2
(c) 18 SCCM 14 800 PPM
Figure 3.32: Ethylene in Nitrogen.
he effect of adding different quantities of
ethylene, as indicated in the sub-figures’
labels, to an inert anode flow rate of
1.2 SLM nitrogen on the deactivation dynamics for MEA PtB CPL under constant
current load. The current loads are specified
in the sub-figures’ legends. The cathode flow
rate was 0.8 SLM oxygen.
77
0 10 20 30 40 50 60 70 80
Current Density (mA/cm2
)
0.0
0.2
0.4
0.6
0.8
1.0
V
olta
g
e (V)
0.0
1.5
3.0
4.5
6.0
7.5
9.0
10.5
P
o
w
er (m
W/c
m2
)
Figure 3.33: Polarization curve for MEA PtB CPL operating on an anode flow rate of
1.2 SLM propane and 18 SCCM ethylene, and a cathode flow rate of 0.8 SLM oxygen. The
polarization curve is parameterized in time, to demonstrate the partial deactivation dynamics at each current density value.
be generating power by utilizing at least some of the propane in the anode flow rate, not just
ethylene. They reason that because increasing the ethylene concentration in the propane
flow rate did not increase the cell’s power density, propane must be reacting, at least once
the cell stabilizes. These results are reproduced here in 3.34.a. Furthermore, they found
that cells running on a propane-ethylene mixture produced more power than cells running
on pure ethylene. This further supported their claim that propane is contributing to the
reaction in the propane-ethylene cells. We conducted a similar set of experiments to theirs
with the sole difference of replacing the propane fuel flow rate with nitrogen. The results,
shown in Figure 3.34.b, indicate that while the cell operating on propane and ethylene has
higher power densities during the transient dynamics phase, both cells perform similarly after
approximately 200 s. The steady-state power density is 10 mW/cm2
for the propane-ethylene
mixture and 8 mW/cm2
for the nitrogen-ethylene mixture.
The discrepancy in steady-state power densities observed in Figure 3.34.a may be attributed to the presence of propane. If we accept this explanation, it implies that propane
actively participates in power generation within the cell. Another consideration is that the
quality of the MEA heavily depends on the fabrication technique employed by the researcher.
78
(a) Ethylene in propane, from [59]
10
0 10
1 10
2 10
3
Time (s)
0
5
10
15
20
25
30
P
o
w
er D
e
nsity (m
W/c
m2
)
416 ppm
624 ppm
832 ppm
1663 ppm
2079 ppm
(b) Ethylene in nitrogen
Figure 3.34: Comparison of our results with previous work on the impact of ethylene additives
on direct propane PEMFC. (a) Results reproduced from [59] demonstrating the influence of
ethylene concentration on power generation at 36 mA/cm2
(for 0 PPM, the cell was initially
activated with a 5 s exposure to 2000 PPM of ethylene in the fuel stream). They employed
a direct propane PEMFC similar to ours, operating at 80 ◦C with an anode flow rate of
1.2 SLM research-grade propane. (b) Our results from the MEA PtB CPL illustrating the
impact of ethylene concentration on deactivation dynamics for a cell operating at 36 mA/cm2
with an anode flow rate of 1.2 SLM nitrogen. The nitrogen is inert and does not contribute
to power production.
Therefore, the MEAs used by Kong et al. and ours may exhibit slightly different qualities,
potentially leading to variations in power density. However, comparing our results in Figures 3.31 and 3.32 reveals that for the same MEA, substituting propane with nitrogen in the
anode flow did not alter performance. This suggests that after an initial transient period,
propane likely does not contribute to power generation, possibly due to its reaction being
fully deactivated over time.
To understand the findings of Kong et al., who observed superior performance in propaneethylene PEMFC compared to those running on pure ethylene, we conducted an experiment
contrasting the performance of a PEMFC fueled by pure ethylene against one utilizing diluted
ethylene in nitrogen. The outcomes of this experiment are shown in Figure 3.35. Our
results demonstrate that diluted ethylene produced higher power output than pure ethylene,
suggesting that the enhanced performance observed by Kong et al. for propane-ethylene
compared to pure ethylene PEMFC might be attributed to the dilution of ethylene rather
than the contribution from propane reactions. Further investigation is necessary to fully
79
10
0 10
1 10
2 10
3
Time (s)
0.0
2.5
5.0
7.5
10.0
12.5
15.0
17.5
P
o
w
er D
e
nsity (m
W/c
m2
)
Diluted ethylene
Non-diluted ethylene
Figure 3.35: Effect of ethylene dilution on the deactivation dynamics of MEA PtB CPL
at 85 ◦C. The deactivation was assessed under two scenarios: non-diluted ethylene with
an anode flow rate of 1.2 SLM of pure ethylene, and diluted ethylene, consisting of a 1.5%
ethylene in nitrogen mixture. A constant current load of 36 mA/cm2 was maintained during
deactivation. Note the logarithmic time-scale.
understand the impact of dilution on the power production levels of both ethylene and
propane PEMFC.
3.6.2 Effect of hydrogen as a fuel additive
Hydrogen is commonly used as fuel in most PEMFC applications due to its high reactivity,
leading to substantial power densities and efficient cell operation [66, 67]. Motivated by
this, we explored the interplay between hydrogen and propane as fuels in our PEMFC to
determine if their interaction is synergistic or antagonistic.
We first investigated the impact of adding small quantities of hydrogen to the direct
propane PEMFC, as done with ethylene in Section 3.6.1, to assess its effectiveness as a fuel
additive. We introduced 18 SCCM (14800 PPM) of hydrogen into an anode flow stream
of 1.2 SLM propane in a cell using the MEA PtC FCS with 1.5 mg/cm2 PtC loading. The
power dynamics results are displayed in Figure 3.36.a. Additionally, we conducted the same
experiment with propane replaced by 1.2 SLM nitrogen, and the outcomes are presented in
Figure 3.36.b.
80
0 5 10 15 20 25 30
Time (s)
0
5
10
15
20
P
o
w
er D
e
nsity (m
W/c
m2
)
10 mA/cm2
15 mA/cm2
20 mA/cm2
30 mA/cm2
(a) Hydrogen in propane
0 2 4 6 8 10 12
Time (s)
0
5
10
15
20
P
o
w
er D
e
nsity (m
W/c
m2
)
10 mA/cm2
15 mA/cm2
20 mA/cm2
30 mA/cm2
(b) Hydrogen in nitrogen
Figure 3.36: The effect of adding 18 SCCM of hydrogen to an anode flow rate of 1.2 SLM
of (a) propane or (b) nitrogen on the power dynamics for MEA PtC FCS with 1.5 mg/cm2
PtC loading under constant current loads. The current loads are specified in the legends.
The cathode flow rate was 0.8 SLM oxygen.
Incorporating a small amount of hydrogen into the PEMFC, whether in a propane or
nitrogen anode flow stream, enhanced the OCV and power densities across all tested current
loads. Notably, hydrogen in nitrogen produced comparable or, in some instances, greater
power than hydrogen in propane. These findings suggest that hydrogen’s high reactivity is
the primary factor determining power density levels. The marginally superior performance of
hydrogen in nitrogen implies that propane may be deactivating some reaction sites, thereby
hindering hydrogen’s reactivity. No extinction was observed in any experiment, eliminating
the observation of deactivation dynamics. Similar patterns and outcomes were noted in
experiments with 3 SCCM hydrogen (2500 PPM), though these results are not included
here.
Given that the PEMFC operating with small amounts of hydrogen did not exhibit deactivation dynamics, we were able to perform polarization curve tests. We conducted experiments to investigate whether propane impedes the performance of a fuel cell operating
with hydrogen. We measured polarization curves for the PEMFC with anode flows consisting of hydrogen and nitrogen, and hydrogen and research-grade propane, at various ratios. The results for the MEA PtC FCS and MEA PtB FCS are displayed in Figures 3.37
and 3.38, respectively, with the MEA PtC FCS having a 1.5 mg/cm2 PtC loading, and the
81
MEA PtB FCS a 4 mg/cm2 PtB loading.
The MEA PtC FCS results (Figure 3.37) show that at high hydrogen concentrations,
mixing hydrogen with either propane or nitrogen did not affect cell performance. However,
reducing the hydrogen concentration to 16% led to better performance in the nitrogenhydrogen cell compared to the propane-hydrogen cell. A general performance decrease was
observed with lower hydrogen concentrations, more so in the propane-hydrogen cell, suggesting that propane might hinder hydrogen’s performance by deactivating certain reaction
sites, making them mechanically or chemically unreachable by the hydrogen fuel. This effect
was more pronounced at lower hydrogen concentrations.
Conversely, the MEA PtB FCS results (Figure 3.38) show no significant performance
difference between the two anode flow mixtures. While reducing hydrogen concentration
generally decreased performance, there was no significant difference between the propanehydrogen and nitrogen-hydrogen cells, possibly due to the higher PtB loading compared to
PtC in the tested MEAs. The difference in loading significantly affected the maximum power
generated by the 100% hydrogen mixture in both MEA types. However, it is noteworthy
that PtC has a higher catalyst utilization rate than PtB.
These findings suggest that hydrogen’s high reactivity predominantly determines cell
performance and dynamics. The interaction between hydrogen and propane in the fuel cell
appears neutral or antagonistic, with no synergistic effects observed. Consequently, adding
small amounts of hydrogen to a direct propane PEMFC does not enhance the utilization
of propane fuel in the cell, leading us to conclude that it does not affect propane reaction
deactivation dynamics.
82
0 100 200 300
Current Density (mA/cm2
)
0.0
0.2
0.4
0.6
0.8
1.0
V
olta
g
e (V)
100% H2, 0% N2
33% H2, 67% N2
16% H2, 84% N2
100% H2, 0% C3H8
33% H2, 67% C3H8
16% H2, 84% C3H8
0
25
50
75
100
125
150
175
P
o
w
er (m
W/c
m2
)
Figure 3.37: Polarization curve for the MEA PtC FCS with 1.5 mg/cm2 PtC loading operating with an anode flow mixture of hydrogen-nitrogen or hydrogen-propane with different
mixing ratios. The mixing ratios are specified in the legend. The cathode flow rate was
0.8 SLM oxygen.
0 100 200 300 400 500
Current Density (mA/cm2
)
0.0
0.2
0.4
0.6
0.8
1.0
V
olta
g
e (V)
100% H2, 0% N2
33% H2, 67% N2
16% H2, 84% N2
100% H2, 0% C3H8
33% H2, 67% C3H8
16% H2, 84% C3H8
0
50
100
150
200
250
300
P
o
w
er (m
W/c
m2
)
Figure 3.38: Polarization curve for the MEA PtB FCS with 4 mg/cm2 PtB loading operating
with an anode flow mixture of hydrogen-nitrogen or hydrogen-propane with different mixing
ratios. The mixing ratios are specified in the legend. The cathode flow rate was 0.8 SLM
oxygen.
83
Chapter 4
Control of direct propane proton exchange membrane fuel cells
4.1 Introduction
Practical PEMFC systems, typically powered by hydrogen, utilize various control strategies
to optimize performance, prevent degradation, and meet power and load demands [68].
This is crucial in fuel cell stacks, where multiple PEMFC are combined to enhance system
power, either in series for higher voltage or parallel for increased current. Key subsystems
requiring control include anode and cathode flow rates, thermal management, humidity and
water management, and power management systems. Moreover, a specific control strategy
is necessary at startup to initiate quickly and avoid “cold-start” phenomena [69].
Reactant flow controllers ensure the cell meets changing external load requirements and
prevent fuel starvation [68], typically using PID feedback controllers [70]. The thermal
management system regulates cell and stack temperatures. As reactant flow rates and external loads vary, so do cell reaction rates and resultant heat, necessitating optimal temperature settings. This system indirectly influences membrane hydration, affecting conductivity.
Closed-loop feedback control systems are common in PEMFC thermal management [71].
The water management system, crucial for maintaining optimal membrane conductivity,
84
manages water produced by the fuel cell reaction. Excessive liquid water can flood the cell,
undermining performance [72]. It typically adjusts cell humidity by controlling the humidity
level of incoming reactant flows, using relative humidity sensors and either classical feedback
or advanced techniques like fuzzy logic controllers [73]. Stationary PEMFC systems may
also recycle water to reduce usage.
The power management system adjusts fuel cell load to align with external requirements. Although various strategies optimize hydrogen fuel cell power performance, hydrogen PEMFC operation is generally stable, lacking the deactivation dynamics seen in
direct propane PEMFCs [68]. Typically, there’s a steady-state optimal operation point for
maximizing power output. The absence of rapid deactivation dynamics narrows the power
management system’s role to matching external load demands and preventing overloading
[74].
Given the significant deactivation dynamics at high current values in direct propane
PEMFC, maintaining a steady-state operation point is impractical, as the cell eventually
fully deactivates. These deactivation and activation dynamics present an opportunity for
implementing active control strategies to enhance cell performance. Motivated by this, we
developed a dynamic model capturing these observed dynamics and explored optimal control
strategies for the cell.
4.2 Dynamic model of the fuel cell
We developed a simplified, physically-based model to capture the dynamics of the direct
propane PEMFC, focusing on the observed experimental results that show deactivation dynamics dependent on current loads and relatively faster activation dynamics in the absence
or reduction of loads [59]. The model aims to capture cell dynamics as a function of current
loads. While cell temperature, humidity level, catalyst loading, and other factors indeed
affect dynamics, this model primarily focuses on current’s impact to simplify and due to its
85
pronounced effect.
In our model, reaction sites within the cell are binary: active or inactive, their fraction
denoted as Xa and Xi
, respectively. Active sites are assumed to produce power proportional
to their fraction. The sum of both fractions is constant over time is equal to unity:
Xa + Xi = 1 (4.1)
We assume that the activation rate follow the law of mass action. This means that the
rate at which inactive reaction sites are turned into active sites depends on the fraction of
inactive sites:
dXi
dt = −K1Xi (4.2)
Where K1 is constant. For a fully deactivated cell, activation is initially rapid upon
removing the current load, then slows as more sites become active.
Further, we assume that the deactivation rate depends on the fraction of active sites
present and and additional term related to the fraction of inactive sites already present:
dXa
dt = −K2Xa − K3X
b
i
(4.3)
Where K2, K3, and b are constants. The −K2Xa term follows from the law of mass action.
The term −K3Xb
i
is needed to capture the sudden onset of deactivation that was observed
experimentally. It captures the phenomena we observed where as the cell is deactivating,
the rate of deactivation is accelerated. Here we assume that this acceleration is dependant
on the the fraction of already existing inactive sites. Without the −K3Xb
i
term, the cell
behaviour reaches a steady state in the long-term given by:
Xa,eq =
K1
K1 + K2
(4.4)
86
for any given current value. This outcome contradicts the experimental observations.
The accelerated rate of activation can be physically explained by a phenomenon where an
inactive site acts as a “seed” site, enabling the formation of chains, such as polymer chains,
which can deactivate adjacent sites. The constant b encapsulates this phenomenon, where
b can be interpreted as the number of inactive sites required to initiate this chain reaction.
For values of b > 1, the model’s predictions align more closely with experimental findings.
Based on those assumption, we propose the following model that captures the direct
propane PEMFC dynamics:
dXa
dt = K1 − (K1 + K2)Xa − K3(1 − Xa)
b
(4.5)
The model solutions are valid in the range 0 < Xa < 1. For values of b = 2, a simple
explicit solution to model can be obtained. For other values of b, numerical integration
methods can be employed. For the rest of this discussion we assume b = 2. Different
values of b were examined numerically and found to not alter the model trends significantly.
Additionally, this model formulation assumes that the state of the system depends only on
the current state, and not on past states, which means that we assume the system has no
”momentum”, as indicated by the absence of higher-order temporal derivatives.
The constants K1, K2, and K3 govern the activation and deactivation rates. Their
values are influenced by factors such as current load, the MEA used, catalyst loading, cell
temperature, humidity level, and reactants’ flow rates. However, for a specific PEMFC and
MEA, with fixed cell temperature, humidity levels, and reactants’ flow rates, these constants
depend solely on the current load. To derive numerical results for our model, we determined
the relationship of these constants with current load for the MEA PtB CPL operating at
85 ◦C with flow rates of 1.2 SLM propane and 0.8 SLM oxygen for the anode and cathode,
respectively.
K1 is governs the activation rate. As activation predominantly occurs in the absence
of current load, we designated K1 as independent of current. For cells initiating activation
87
immediately after reaching full deactivation, they typically regain 75% of their maximum
OCV within a time constant of 2 s, as shown in Fig. 3.23. Based on this, we set K1 = 0.5 s−1
.
With K1 established, K2 and K3 dictate the dynamics under current loads, particularly
deactivation dynamics. To ascertain how K2 and K3 vary with current load, we analyzed
deactivation dynamics data from operating the cell under seven different constant current
loads ranging from 5 mA/cm2
to 36 mA/cm2
. For each load, we manually selected values for
K2 and K3 to enhance the model’s congruence with experimental findings, assuming a linear
correlation between cell voltage and the predicted fraction of active sites, Xa. The choice of
K2 and K3 aimed to match experimental trends, especially concerning power levels before
complete deactivation and the duration until extinction.
Based on the manually selected values of K2 and K3, and the corresponding current
density values, we formulated the following functional relationship:
Ki = ai +
bi
1 + ci exp(−diI)
(4.6)
Where i = 1, 2, 3 and I is the current density in mA/cm2
. Eq. 4.6 represents an asymmetric sigmoid function. A visualization of a simple sigmoid function, sigmoid(x) = 1
1+exp (x)
,
is shown in Figure 4.1. This sigmoid form was selected because it accurately captures the
cell’s behavior at lower current values where no extinction occurs, and at higher current
values where extinction is observed.
To determine the values of ai
, di
, ci
, and di
for i = 1, 2, we conducted a joint optimization
to align the manually selected K values with the corresponding current densities. This
optimization utilized the least-squares algorithm from the Python library scipy.optimize
[75], with the results displayed in Table 4.1. We set d2 = d3 to ensure the direct current
dependence is the same for both K2 and K3. Additionally, Eq. 4.6 is also applied to describe
K1, with a1 = 0.5 s−1 and b1 = 0. These constant values are valid for current densities up to
36 mA/cm2
, as our data did not include higher densities.
With these values of Ki
, the model closely matches experimental observations, accurately
88
10.0 7.5 5.0 2.5 0.0 2.5 5.0 7.5 10.0
x
0.0
0.2
0.4
0.6
0.8
1.0
sigmoid(x)
Figure 4.1: Visualization of a symmetric sigmoid function, sigmoid(x) = 1
1+exp (x)
.
Rate constant a (s−1
) b (s−1
) c (-) d (cm2/mA)
K1 0.50 0 N/A N/A
K2 0.11 0.14 8.98 0.23
K3 0.35 0.22 10.64 0.23
Table 4.1: Parameters for sigmoid-fit (Eq. 4.6) to rate constants for PEMFC dynamic model
(Eq. 4.5).
predicting deactivation and activation dynamics. This is demonstrated by comparing model
predictions with experimental results in Figure 4.2. The model was solved numerically in
simulink [76]. In both scenarios, a current load of I = 30 mA/cm2 was applied until full
deactivation occurred (V < 0.1 or Xa < 0.1), then the current was removed by setting
I = 0 mA/cm2
. For the model, we monitored the predicted value of Xa, whereas for the
experiment, we observed the cell voltage. Setting the model initial value Xa,0 = 0.8 improved
alignment with experiment observations. The results show that the model is in good agreement with the experimental data, successfully capturing both deactivation and activation
dynamics.
89
0 20 40 60 80 100
Time (s)
0.0
0.2
0.4
0.6
0.8
Xa pre
dictio
n
(a) Model
0 20 40 60 80 100
Time (s)
0.0
0.2
0.4
0.6
0.8
V
olta
g
e (V)
(b) Experiment
Figure 4.2: A comparison of the dynamic propane PEMFC model and the experimental
results. For both results, a current load of I = 30 mA/cm2 was applied on the cell until
full deactivation was reached(V < 0.1 or Xa < 0.1), then the current load was removed, by
setting I = 0 mA/cm2
. (a) Shows the model prediction for Xa. We set Xa,0 = 0.8 to better
match experimental results. (b) Shows the experimental voltage dynamics as the cell goes
through one deactivation-activation cycle.
4.3 Formulating the control problem
While the dynamics of the fuel cell depend on various variables, for a specific MEA and set
of operational parameters, the dynamics can essentially be modeled as a function of current
density alone. This simplification enables us to outline a straightforward control problem,
defining the system state, input, and output, respectively, as follows:
x(t) = V (t) (4.7)
u(t) = I(t) (4.8)
y(t) = P(t) = I(t)V (t) (4.9)
where V (t) represents the cell voltage, I(t) the applied current density, and P(t) the cell’s
power density. The system state-space model is then:
90
x, y PEMFC
Dynamics
u Current
Controller
Figure 4.3: General feedback control architecture for the direct propane PEMFC.
x˙ = g(x(t), u(t)) (4.10)
y = x(t)u(t) (4.11)
Using the dynamics model from the previous section, we can expand Eq. 4.10, assuming
Xa(t) = V (t), a reasonable approximation given the correlation observed in Figure 4.2. This
relation may vary across different cells but is likely still linear. For b = 2, Eq. 4.10 becomes:
x˙ = f1(u) − (f1(u) + f2(u))x − f3(u)(1 − x)
2
(4.12)
fi = ai +
bi
1 + ci exp(−diu)
(4.13)
For our specific PEMFC, the parameters ai
, bi
, ci
, and di adhere to the values listed in
Table 4.1. Note that the system dynamics are non-linear because of the (1 − x)
2
term and
the sigmoid function in Eq. 4.13.
The ensuing control challenge is to design a controller that can regulate the power output
level of the direct propane PEMFC by manipulating the current density load on the cell.
Figure 4.3 depicts a schematic of a typical feedback control system architecture.
91
4.4 On-off control
Kong et al. proposed a “load-interrupt” operation mode, as depicted in Figure 3.3 [59],
where the current is applied for 20 s and then removed for 5 s. This cycle leverages the slow
deactivation and rapid activation dynamics of the cell to enhance average power output. The
“load-interrupt” mode essentially functions as a feed-forward On-off controller, relying on
heuristics without utilizing state feedback, which may not optimize performance or guarantee
stability. Insufficient off-time may prevent adequate reactivation, reducing average power
density over time. Thus, we explored a feedback on-off controller.
On-off controllers, a non-linear control architecture, toggle the control action based on
whether the process variable surpasses or drops below a set of thresholds, providing a straightforward mechanism for regulation. The design for an on-off controller using the cell’s voltage
for feedback is illustrated in Figure 4.4.a, with its governing control law depicted in Figure
4.4.b. This law outlines an on-off controller with hysteresis, switching the current between
high (Ih) and low (Il) values at predetermined voltage levels (Vh and Vl). A proportional
current is applied within specific voltage ranges. For the rest of the discussion, we set Va = Vl
and Vb = Vh, which reduces the control law to an on-off controller with hysteresis and no
proportional control effort.
Utilizing the dynamic model proposed earlier, we simulated the on-off controller’s performance in simulink. Simulations offer a quick way to test various control architectures,
especially given the time-consuming nature of experimental PEMFC testing. Simulation
results, as shown in 4.5, indicate cyclic stability after the initial cycle. Due to faster activation compared to deactivation dynamics, the “off” period is notably shorter than the
“on” period. For this simulation, we set Ih = 30 mA/cm2
, Il = 0 mA/cm2
, Vl = Va = 0.3 V,
and Vh = Vb = 0.6 V. The simulation results indicate that an average power density of
8.8 mW/cm2
is achievable with the on-off controller with the given parameters.
Following the simulation insights, we deployed the on-off controller on our direct PEMFC,
92
x
PEMFC
Dynamics
u On-off Current
Controller
y
(a) On-off control architecture (b) On-off control law
Figure 4.4: The design of an On-off current controller with hysteresis.
0 50 100 150
time (s)
0
10
20
30
C
u
r
r
e
n
t
d
e
n
sit
y (m
A
/
c
m
2
)
0 50 100 150
time (s)
0.2
0.4
0.6
0.8
V
olt
a
g
e (V)
0 50 100 150
time (s)
0
5
10
15
20
25
P
o
w
e
r
d
e
n
sit
y (m
W/
c
m
2
)
Instant Power
Average Power
Figure 4.5: Simulation of an on-off controller performance using the architecture from figure
4.4. We set Ih = 30 mA/cm2
, Il = 0 mA/cm2
, Vl = Va = 0.3 V, and Vh = Vb = 0.6 V.
93
focusing on varying control law parameters using the MEA PtB CPL. Figure 4.6 presents
results from implementing a controller with settings Ih = 30 mA/cm2
, Il = 0 mA/cm2
, Vl =
0.1 V, and Vh = 0.7 V. In Figure 4.6.a, the voltage dynamics demonstrate the controller’s
ability to sustain the cell voltage within specified limits, Vl and Vh, for 1000 s. Figure 4.6.b
shows the instantaneous and average power output, with power toggling between high and
low modes as the current switches. The average power, calculated over 100 s intervals,
approximates 5 mW/cm2
. Figure 4.6.c details the on-off cycle duration, defined by the time
for a complete on-off sequence, determined from consecutive peaks in instantaneous power.
The cycle duration initially exceeds 25 s, then reduces and stabilizes below 10 s within the first
few cycles. This observation deviates from the model’s prediction of stable cycle durations,
as seen in Figure 4.5. The discrepancy may arise from inaccuracies in assuming a linear
correlation between Xa and cell voltage, suggesting that the cell’s cycle stability under our
controller might not directly translate to constant active site fractions. Alternatively, it’s
possible that the rates of activation and deactivation involve higher-order temporal dynamics
not captured in our model.
Additionally, we evaluated the on-off controller under varying control settings. Figure 4.7
displays outcomes from three distinct experimental sets. The controller in Figures 4.7.a-b
and 4.7.c-d mirror those in Figure 4.6 but with adjusted Ih values. In Figures 4.7.a-b, Ih was
increased to 40 mA/cm2
, increasing deactivation rates and reducing the on-off cycle duration
to 6 s. This adjustment led to higher peak instantaneous power without altering the average
power output. Conversely, Figures 4.7.c-d show the effects of reducing Ih to 25 mA/cm2
,
which, as anticipated, lengthened the cycle duration to 10 s due to a slower deactivation
rate. Although the peak power decreased, the average power output stayed consistent.
In Figures 4.7.e-f, the controller was set as in Figure 4.6, but with Il adjusted to
2 mA/cm2
. This change keeps power generation active during off periods but decelerates
the cell’s reactivation process. This is demonstrated by the increased cycle duration to
approximately 17 s. However, the average power output increased to 5.4 mW/cm2
.
94
0 200 400 600 800 1000
Time (s)
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
V
olta
g
e (V)
(a) Voltage dynamics
0 200 400 600 800 1000
Time (s)
0
5
10
15
20
P
o
w
er D
e
nsity (m
W/c
m2
)
Instant Power
Average Power
(b) Power dynamics
0 200 400 600 800 1000
Time (s)
10.0
12.5
15.0
17.5
20.0
22.5
25.0
Period (s)
(c) On-off period duration
Figure 4.6: Experimental results for implementing the on-off controller. We set Ih =
30 mA/cm2
, Il = 0 mA/cm2
, Vl = 0.1 V, and Vh = 0.7 V. (a) Shows the voltage dynamics, (b) shows the power dynamics, and (c) Shows the the period for one on-off cycle.
95
0 100 200 300 400 500
Time (s)
0
5
10
15
20
25
P
o
w
er D
e
nsity (m
W/c
m2
)
Instant Power
Average Power
(a) Power dynamics, Ih = 40 mA/cm2
0 100 200 300 400 500
Time (s)
6
8
10
12
Period (s)
(b) On-off period duration
0 20 40 60 80 100 120 140
Time (s)
0.0
2.5
5.0
7.5
10.0
12.5
15.0
17.5
P
o
w
er D
e
nsity (m
W/c
m2
)
Instant Power
Average Power
(c) Power dynamics, Ih = 25 mA/cm2
20 40 60 80 100 120
Time (s)
10.0
10.5
11.0
11.5
Period (s)
(d) On-off period duration
0 200 400 600 800 1000 1200 1400
Time (s)
0.0
2.5
5.0
7.5
10.0
12.5
15.0
P
o
w
er D
e
nsity (m
W/c
m2
)
Instant Power
Average Power
(e) Power dynamics, Il = 2 mA/cm2
0 200 400 600 800 1000 1200 1400
Time (s)
15.0
17.5
20.0
22.5
25.0
27.5
30.0
32.5
Period (s)
(f) On-off period duration
Figure 4.7: Experimental results for implementing the on-off controller. For all experiments,
we set Vl = 0.1 V, and Vh = 0.7 V. For (a, b), Ih = 40 mA/cm2
, Il = 0 mA/cm2
. For (c, d)
Ih = 25 mA/cm2
, Il = 0 mA/cm2
. For (e, f), Ih = 30 mA/cm2
, Il = 2 mA/cm2
.
96
0 50 100 150 200
Time (s)
0
10
20
30
P
o
w
er D
e
nsity (m
W/c
m2
)
Instant Power
Average Power
Figure 4.8: Experimental results for implementing the on-off controller on a high temperature
cell operating at 101 ◦C. We set Ih = 55 mA/cm2
, Il = 0 mA/cm2
, Vl = 0.1 V, and Vh =
0.7 V.
To evaluate the benefits of using an on-off controller, we compared the cell’s performance with the controller to its operation in constant current mode. For the same MEA
(MEA PtC CPL), constant current operation led to extinction at currents greater than
15 mA/cm2
, as shown in Figure 3.13. The maximum current that allowed for continuous
operation without extinction was 10 mA/cm2
, at which the cell maintained an average power
density of 3.4 mW/cm2
. In contrast, the most effective on-off controller setup we evaluated
(Figure 4.7e,d) achieved a steady-state average power density of 5.4 mW/cm2
, indicating a
59% improvement in power density over constant current operation.
Further experiments were conducted with the on-off controller on a high-temperature
cell utilizing MEA PtB CPL and operating at 101 ◦C. These results are shown in Figure
4.8. At this higher temperature, we could operate the cell with high “on” current values of
Ih = 55 mW/cm2
. The cycle period averaged 10 s, achieving an average power density of
11 mW/cm2
. Operating at higher current loads and increased temperature with the on-off
controller resulted in greater average power output. Compared to the high-temperature cell’s
performance in constant current mode (Figure 3.19), the use of the on-off controller showed
an approximate 37% increase in average power.
97
Figure 4.9: Simplified block diagram of a MPC-based control loop [77].
4.5 Model predictive control
Model Predictive Control (MPC) represents a sophisticated approach to control systems,
relying on predictive models to optimize control actions over a finite time horizon [77].
Unlike conventional controllers, which typically operate based on current system states,
MPC considers future system behavior by iteratively solving an optimization problem to
determine the optimal control input sequence (fig. 4.9). This enables MPC to account for
constraints, disturbances, and dynamic behavior, making it particularly suited for complex
and nonlinear systems where traditional control methods may struggle to maintain stability
or performance. Crucially, MPC requires a dynamic model of the system, which accurately
represents its behavior under various conditions. This model serves as the foundation for
predicting future system responses and optimizing control actions accordingly. Therefore,
one of the primary reasons for utilizing MPC is the capability to leverage the predictive
capabilities of a well-tuned model.
Furthermore, nonlinear MPC extends this strategy to systems with nonlinear dynamics [78]. While the implementation of nonlinear MPC introduces additional computational
challenges due to the increased complexity of solving optimization problems with nonlinear
constraints and objectives, it improves the robustness of the controller. The shift from traditional control strategies to MPC involves a change in the control engineer’s focus: from
designing control laws to designing predictive models that accurately describe system be98
havior. Once a robust model is established, the MPC algorithm optimizes control actions by
formulating and solving an optimization problem, providing an optimal control law for the
specific dynamics of the system.
Since the direct propane PEMFC dynamical model we developed in section 4.2 is able to
predict observed experimental results (fig. 4.2), we investigated the potential of designing an
MPC to improve the performance of the cell. The MPC problem can generally be formulated
as follows:
minimize
u(·)
J(x(t0), u(·)) (4.14a)
subject to ˙x(t) = f(x(t), u(t), t), (4.14b)
x(t0) = x0, (4.14c)
x(t) ∈ X , ∀t ∈ [t0, tf ], (4.14d)
u(t) ∈ U, ∀t ∈ [t0, tf ], (4.14e)
g(x(t), u(t), t) ≤ 0, (inequality constraints) (4.14f)
h(x(t), u(t), t) = 0, (equality constraints) (4.14g)
where:
• x(t) is the state vector of the system at time t.
• u(·) is the control input function over the time horizon.
• J is the cost function to be minimized by the MPC.
• f(x(t), u(t), t) is the system dynamics, which can be non-linear.
• x0 is the initial condition of the state vector.
• X and U are the feasible sets for the states and control inputs, respectively.
99
• g(x(t), u(t), t) ≤ 0 are the inequality constraints that must be satisfied by the states
and controls at all times.
• h(x(t), u(t), t) = 0 are the equality constraints that must be satisfied by the states and
controls at all times.
For the direct propane PEMFC, utilizing the problem definition presented in Section 4.3,
the MPC problem over a specified time horizon can be expressed as follows:
minimize
u(·)
J =
Z tf
t0
1
1 + x(t)u(t)
dt (4.15a)
subject to ˙x(t) = f1(u) − (f1(u) + f2(u))x − f3(u)(1 − x)
2
, (4.15b)
fi(u) = ai +
bi
1 + ci exp(−diu)
(4.15c)
x(t0) = x0, (4.15d)
0 ≤ x(t) ≤ 1, (4.15e)
0 ≤ u(t) ≤ Imax, (4.15f)
|u˙(t)| ≤
dI
dt
max
(4.15g)
We define the cost function, J, as the integral of the reciprocal of the power density
produced over the time horizon. We impose constraints on the allowable cell voltage, which
must remain between 0 and 1 V, and on the maximum allowable current density, set to
Imax = 36 mA/cm2
. Additionally, we limit the maximum rate of change of current density
to
dI
dt
max
= 5 mA/(cm2
s) to avoid control chatter.
Using the Simulink platform and the open-source nonlinear optimization library CasADi
[79], we conducted simulations to evaluate the performance of the MPC algorithm applied
to our dynamical model. The implementation of the MPC was configured with a time
horizon of 100 s and a time step of 0.25 s. The MPC algorithm determined the optimal
10
0 50 100 150
time (s)
10
20
30
C
u
r
r
e
n
t
d
e
n
sit
y (m
A
/
c
m
2
)
0 50 100 150
time (s)
0.2
0.4
0.6
0.8
V
olt
a
g
e (V)
0 50 100 150
time (s)
5
10
15
20
25
P
o
w
e
r
d
e
n
sit
y (m
W/
c
m
2
)
Instant Power
Average Power
Figure 4.10: Simulation of the MPC controller performance using formulation in Eq. 4.15.
control sequence within the specified horizon, applying only the immediate control effort
and disregarding subsequent steps. It is important to note the inherent trade-off between
the duration of the time horizon and the optimality of the solution; longer horizons enhance
the optimality of the control effort at the expense of increased computational demands. The
initial condition was set as x0 = 0.8 V. Figure 4.10 presents the simulation outcomes.
The simulation results indicate that the MPC algorithm converges to a control strategy
that resemble the on-off control strategy introduced in the preceding section. Without us
predetermining a specific control approach, the MPC modulated the current load between
Imax and 10 mA/cm2
to maintain the cell voltage within the range of 0.35 V to 0.4 V, while
respecting the maximum rate of current change
dI
dt
max
. This outcome suggests that an onoff strategy, incorporating a non-zero “off” current load, approaches optimal control. The
simulated MPC controller achieved an average power output of 12.1 mW/cm2
, indicating
a 35% improvement over the simulated on-off controller results (figure 4.5). The primary
advantage of MPC over conventional on-off control lies in its robustness to disturbances and
its adaptability to temporal variations in cell dynamics.
To demonstrate the robustness of the MPC controller, we introduced uniform noise with
a magnitude range of [0, 0.15] to the Ki constants within the dynamical model, with a
sampling frequency equal to the simulation solution frequency. Concurrently, the predictive
model employed by the MPC adhered to the dynamics outlined in Eq. 4.15, without the
added noise. Figure 4.11 presents the outcomes of this simulation. Despite discrepancies
10
0 50 100 150
time (s)
10
20
30
C
u
r
r
e
n
t
d
e
n
sit
y (m
A
/
c
m
2
)
0 50 100 150
time (s)
0.2
0.4
0.6
0.8
V
olt
a
g
e (V)
0 50 100 150
time (s)
5
10
15
20
P
o
w
e
r
d
e
n
sit
y (m
W/
c
m
2
)
Instant Power
Average Power
Figure 4.11: Simulation of the MPC Controller robustness in a PEMFC dynamical system
with noise-perturbed rate constants (Ki).
between the actual cell dynamics and the model utilized by the MPC, the control strategy
executed by the MPC appeared to be robust and optimally tuned, leading to to sustained
average power production. Specifically, the MPC produced an average power density of
10.3 mW/cm2
. This output is 15% lower than that achieved by the MPC in simulations
without model disturbances, yet 16% higher than the performance of the simulated on-off
controller (figure 4.5). This outcome suggests that while discrepancies between the MPC
model and actual system dynamics might reduce the average power density output, the
controller remains robust.
While simulations of the MPC controller with long time horizons and high sampling
rates are feasible, implementing these controllers experimentally presents significant challenges due to the substantial real-time computational load required. Strategies to mitigate
computational demands involves either reducing the sampling rate or shortening the MPC
horizon. However, reducing the sampling rate, which effectively means increasing the MPC
solution’s time step, is constrained by the activation dynamics’ small timescale compared
to the slower deactivation dynamics. To effectively capture the activation dynamics, we are
bounded by a maximum time step of approximately 0.5 s. Moreover, excessively reducing
the time horizon can decrease the controller’s optimality since it leads to optimization over
a shorter horizon than intended by the MPC’s cost function design (Eq. 4.15a). If the
MPC horizon is smaller than the average duration of one on-off cycle (approximately 15 s),
102
the controller tends to optimize for instant power production, ultimately converging into a
steady-state solution. To allow for the possibility of the MPC converging on a cyclic control
strategy that maximizes average power output, as defined by the cost function, the time
horizon must sufficiently exceed the expected cycle time.
Early experimental implementations of the MPC controller with time horizons of 7.5 s and
15 s, shown in Figure 4.12, confirm these predictions. As anticipated, when the time horizon
does not extend beyond the on-off cycle’s typical length, the controller converges into a
steady-state control solution, which avoids extinction and produces a constant power density
output. Initially, the controller exhibits rapid chatter at higher voltage levels but eventually
stabilizes, maintaining a steady control effort of approximately 12 mA/cm2 at 0.35 V. Our
current research efforts are focused on improving the MPC controller’s implementation,
which would permit employing longer time horizons in practical setups, thereby validating
the simulation results over similar durations.
103
0 50 100 150
Time (s)
0
10
20
30
40
C
u
r
r
e
n
t
d
e
n
sit
y (m
A
/
c
m
2
)
0 50 100 150
Time (s)
0.0
0.2
0.4
0.6
0.8
V
olt
a
g
e (V)
0 50 100 150
Time (s)
0
5
10
15
20
P
o
w
e
r
d
e
n
sit
y (m
W/
c
m
2
)
Instant Power
Average Power
(a) Time-horizon length = 7.5 s, ∆t = 0.25 s
0 100 200 300
Time (s)
0
10
20
30
40
C
u
r
r
e
n
t
d
e
n
sit
y (m
A
/
c
m
2
)
0 100 200 300
Time (s)
0.2
0.4
0.6
0.8
V
olt
a
g
e (V)
0 100 200 300
Time (s)
0
5
10
15
P
o
w
e
r
d
e
n
sit
y (m
W/
c
m
2
)
Instant Power
Average Power
(b) Time-horizon length = 15 s, ∆t = 0.25 s
Figure 4.12: Experimental implementation of the MPC controller with short time horizons,
as specified in the captions of the subfigures. For both controllers, the solution’s time step
was maintained at 0.25 s, corresponding to a sampling rate of 4 Hz.
104
Chapter 5
Future Research Directions
This dissertation has explored innovative pathways for small-scale power generation through
the use of chemical fuels such as hydrogen and propane and the development and characterization of miniature catalytic-combustion engines and direct hydrocarbon proton exchange
membrane fuel cells (PEMFC). The insights gained lay the groundwork for a host of potential
research directions aimed at pushing the boundaries of power-density, efficiency, applicability,
and fundamental understanding of these promising systems.
One of the significant challenges identified in the study of direct propane PEMFC is the
phenomenon of cell deactivation, particularly under high current loads. This issue stands
as a critical barrier to the practical application and broader adoption of these fuel cells
as portable power generation devices. To address this, future work must prioritize efforts
to understand the underlying mechanisms of deactivation. Advanced analytical techniques,
such as in-situ spectroscopy facilitated by PEMFC designed with optical access, offer a
promising pathway to explore the catalyst surface chemistry in detail [80]. Understanding the
molecular chemistry and processes that lead to catalyst deactivation can illuminate strategies
to prevent this phenomenon, thereby enhancing cell power-density and performance.
Moreover, increasing the power outputs of propane PEMFC to levels that are competitive
with or superior to conventional batteries, on a mass-basis, is essential for their success in
portable electronics and other application domains. Achieving this requires not only mitigating deactivation but also exploring new catalysts, membrane materials, and fuel delivery
105
systems that can operate more efficiently and withstand prolonged use. The integration
of such improvements could significantly increase the viability of PEMFC as a leading solution for portable and small-scale power generation, offering energy densities and power
capabilities beyond what current battery technologies can provide.
Beyond these specific challenges, the integration of small-scale power generation devices
into real-world applications remains a crucial next step. This involves the deployment of
miniature catalytic-combustion engines in micro-robotic systems, and the direct propane
PEMFC in portable electronics. Integrating propane PEMFC technology into real-world
applications will require the development of compact and efficient thermal and water management systems, alongside robust fuel delivery mechanisms, all of which still need to be
investigated.
Optimizing the design and operational parameters of miniature catalytic-combustion engines presents an immediate opportunity for future work. While the work presented utilized
hydrogen as the fuel for the miniature engine, an important research goal is to use hydrocarbon fuel, which allow the engine to be used in real micro-robotics applications. Exploring alternative materials with higher catalytic efficiency, refining the engine geometry for improved
heat management, and developing advanced control strategies for fuel delivery systems could
significantly increase the operational frequency and power output of these engines. These
improvements are essential for broadening the range of micro-robotic applications where
these engines can be effectively deployed.
Complementary to experimental and practical application efforts, the advancement of
theoretical models and computational simulations of the processes involved in the small-scale
power generation devices is essential. Such work can offer deeper insights into the operation
dynamics of these devices, assist in the optimization of designs, and predict system behavior
under diverse conditions. Advanced modeling can guide experimental studies and refine
control strategies for both miniature catalytic-combustion engines and propane PEMFC.
The research presented in this dissertation represents a significant advancement in the
106
development of small-scale power generation technologies that run on chemical fuels. By
pursuing the research directions outlined above, it is conceivable that substantial progress
could be made toward realizing highly efficient and reliable power generation devices that
exploit the high energy density of hydrocarbons to deliver small-scale portable power.
107
Bibliography
[1] Jenay M Beer, Arthur D Fisk, and Wendy A Rogers. “Toward a framework for levels of
robot autonomy in human-robot interaction”. In: Journal of human-robot interaction
3.2 (2014), p. 74.
[2] Min-Soo Kim, Hyun-Taek Lee, and Sung-Hoon Ahn. “Laser Controlled 65 Micrometer Long Microrobot Made of Ni-Ti Shape Memory Alloy”. In: Advanced Materials
Technologies 4.12 (2019), p. 1900583.
[3] Seth Hollar et al. “Solar powered 10 mg silicon robot”. In: The Sixteenth Annual International Conference on Micro Electro Mechanical Systems, 2003. MEMS-03 Kyoto.
IEEE. IEEE. 2003, pp. 706–711.
[4] Takashi Yasuda, Isao Shimoyama, and Hirofumi Miura. “Microrobot actuated by a
vibration energy field”. In: Sensors and Actuators A: Physical 43.1-3 (1994), pp. 366–
370.
[5] Karl Vollmers et al. “Wireless resonant magnetic microactuator for untethered mobile
microrobots”. In: Applied Physics Letters 92.14 (2008), p. 144103.
[6] Chytra Pawashe, Steven Floyd, and Metin Sitti. “Modeling and experimental characterization of an untethered magnetic micro-robot”. In: The International Journal of
Robotics Research 28.8 (2009), pp. 1077–1094.
[7] Dongwoo Lee et al. “Design of centimeter-scale inchworm robots with bidirectional
claws”. In: 2011 IEEE International Conference on Robotics and Automation. IEEE.
2011, pp. 3197–3204.
[8] Benjamin Goldberg et al. “Power and control autonomy for high-speed locomotion with
an insect-scale legged robot”. In: IEEE Robotics and Automation Letters 3.2 (2018),
pp. 987–993.
[9] Daniel S Contreras, Daniel S Drew, and Kristofer SJ Pister. “First steps of a millimeterscale walking silicon robot”. In: 2017 19th International Conference on Solid-State
Sensors, Actuators and Microsystems (TRANSDUCERS). IEEE. 2017, pp. 910–913.
[10] Byungkyu Kim et al. “An earthworm-like micro robot using shape memory alloy actuator”. In: Sensors and Actuators A: Physical 125.2 (2006), pp. 429–437.
[11] Aaron M Hoover, Erik Steltz, and Ronald S Fearing. “RoACH: An autonomous 2.4 g
crawling hexapod robot”. In: 2008 IEEE/RSJ International Conference on Intelligent
Robots and Systems. IEEE. 2008, pp. 26–33.
108
[12] Jean-Marc Breguet et al. “A review on actuation principls for few cubic millimeter
sized mobile micro-robots”. In: Proceedings of the 10th International Conference on
New Actuators (Actuator 2006). CONF. 2006, pp. 374–381.
[13] Z Ge Joey, Longlong Chang, and N´estor O P´erez-Arancibia. “Position Control of a
Shape-Memory Alloy Actuator Using a Preisach-Model-Based Inverse-Temperature
Method”. In: 2019 American Control Conference (ACC). IEEE. 2019, pp. 3801–3808.
[14] Darren J Hartl and Dimitris C Lagoudas. “Aerospace applications of shape memory
alloys”. In: Proceedings of the Institution of Mechanical Engineers, Part G: Journal of
Aerospace Engineering 221.4 (2007), pp. 535–552.
[15] Xiaobin Ji et al. “An autonomous untethered fast soft robotic insect driven by lowvoltage dielectric elastomer actuators”. In: Science Robotics 4.37 (2019).
[16] Xuan-Truc Nguyen et al. “SMALLBug: A 30-mg Crawling Robot Driven by a HighFrequency Flexible SMA Microactuator”. In: IEEE Robotics and Automation Letters
5.4 (2020), pp. 6796–6803.
[17] Ghassan Zubi et al. “The lithium-ion battery: State of the art and future perspectives”.
In: Renewable and Sustainable Energy Reviews 89 (2018), pp. 292–308.
[18] Leonardo Ricotti et al. “Biohybrid actuators for robotics: A review of devices actuated
by living cells”. In: Science Robotics 2.12 (2017).
[19] AD Johnson and J Kramer. “State-of-the-art of shape memory actuators”. In: 6th
international conference on new actuators. Bremen, Germany. 1998.
[20] Xiufeng Yang, Longlong Chang, and N´estor O P´erez-Arancibia. “An 88-milligram
insect-scale autonomous crawling robot driven by a catalytic artificial muscle”. In:
Science Robotics 5.45 (2020).
[21] Zhenlong Wang et al. “A micro-robot fish with embedded SMA wire actuated flexible
biomimetic fin”. In: Sensors and Actuators A: Physical 144.2 (2008), pp. 354–360.
[22] Scott Ryan Kirkpatirck et al. Shape memory alloy MEMS heat engine. US Patent
7,444,812. 2008.
[23] Dragan Avirovik et al. “Miniature shape memory alloy heat engine for powering wireless sensor nodes”. In: Energy Harvesting and Systems 1.1-2 (2014), pp. 13–18.
[24] Hiroyuki Aono et al. “Formulation and optimization of pulley-gear-type SMA heat engine toward microfluidic MEMS motor”. In: 2014 IEEE/RSJ International Conference
on Intelligent Robots and Systems. IEEE. 2014, pp. 2839–2845.
109
[25] Von Howard Ebron et al. “Fuel-powered artificial muscles”. In: Science 311.5767 (2006),
pp. 1580–1583.
[26] Yonas Tadesse et al. “Hydrogen-fuel-powered bell segments of biomimetic jellyfish”.
In: Smart Materials and Structures 21.4 (2012), p. 045013.
[27] Michael Wehner et al. In: Nature 536.7617 (2016), pp. 451–455.
[28] Yang Shi, John Whalen, and Paul Ronney. “Hydrocarbon Ignition on High Surface
Area Pt-Electroplated Wires”. In: AIAA Scitech 2019 Forum. 2019, p. 2370.
[29] Fares Maimani et al. “A 7-mg miniature catalytic-combustion engine for millimeterscale robotic actuation”. In: Sensors and Actuators A: Physical 341 (2022), p. 112818.
[30] Z Chaudhry and Craig A Rogers. “Bending and shape control of beams using SMA
actuators”. In: Journal of Intelligent Material Systems and Structures 2.4 (1991),
pp. 581–602.
[31] Gregory Nellis and Sanford A. Klein. Heat transfer. Cambridge University Press, 2012.
[32] CM Spadaccini, J Peck, and IA Waitz. “Catalytic combustion systems for microscale
gas turbine engines”. In: (2007).
[33] Junuthula Narasimha Reddy. Mechanics of laminated composite plates and shells: theory and analysis. CRC press, 2003.
[34] Ne Arancibia, Steve Gibson, Tsu-Chin Tsao, et al. “Adaptive control of MEMS mirrors
for beam steering”. In: ASME 2004 International Mechanical Engineering Congress
and Exposition. American Society of Mechanical Engineers Digital Collection. 2004,
pp. 71–80.
[35] Yong-Lae Park et al. “Design and control of a bio-inspired soft wearable robotic device
for ankle–foot rehabilitation”. In: Bioinspiration & biomimetics 9.1 (2014), p. 016007.
[36] Kambiz Farahmand and Jonathan W Kaufman. “Experimental measurement of fine
thermocouple response time in air”. In: Experimental heat transfer 14.2 (2001), pp. 107–
118.
[37] John William Draper. “On the production of light by heat”. In: Journal of the Franklin
Institute 44.2 (1847), pp. 122–128.
[38] P Cho and Chung King Law. “Catalytic ignition of fuel/oxygen/nitrogen mixtures over
platinum”. In: Combustion and Flame 66.2 (1986), pp. 159–170.
110
[39] James Larminie, Andrew Dicks, and Maurice S McDonald. Fuel cell systems explained.
Vol. 2. J. Wiley Chichester, UK, 2003.
[40] Karl Kordesch and G¨unter Simader. “Fuel cells and their applications”. In: (1996).
[41] Yun Wang et al. “Materials, technological status, and fundamentals of PEM fuel cells–a
review”. In: Materials today 32 (2020), pp. 178–203.
[42] Vladimir S Bagotsky. Fuel cells: problems and solutions. Vol. 56. John Wiley & Sons,
2012.
[43] Sunita Satyapal. “Hydrogen and fuel cell program overview”. In: 2011 Annual Merit
Review and Peer Evaluation Meeting, Fuel Cell Technology Program, Office of Energy
Efficiency and Renewable Energy, US Department of Energy, Washington, DC. 2011.
[44] GK Surya Prakash, Federico A Viva, and George A Olah. “Electrochemical reduction
of CO2 over Sn-Nafion® coated electrode for a fuel-cell-like device”. In: Journal of
Power Sources 223 (2013), pp. 68–73.
[45] Tatiane da Silva Veras, Thiago Simonato Mozer, Aldara da Silva C´esar, et al. “Hydrogen: trends, production and characterization of the main process worldwide”. In:
International journal of hydrogen energy 42.4 (2017), pp. 2018–2033.
[46] Bruno Scrosati. “Power sources for portable electronics and hybrid cars: lithium batteries and fuel cells”. In: The Chemical Record 5.5 (2005), pp. 286–297.
[47] P Linstorm. “NIST chemistry webbook, NIST standard reference database number
69”. In: J. Phys. Chem. Ref. Data, Monograph 9 (1998), pp. 1–1951.
[48] Tolga Taner. “The novel and innovative design with using H2 fuel of PEM fuel cell:
Efficiency of thermodynamic analyze”. In: Fuel 302 (2021), p. 121109.
[49] A Heinzel and VM Barragan. “A review of the state-of-the-art of the methanol crossover
in direct methanol fuel cells”. In: Journal of Power sources 84.1 (1999), pp. 70–74.
[50] LW Niedrach. “The performance of hydrocarbons in ion exchange membrane fuel cells”.
In: Journal of The Electrochemical Society 109.11 (1962), p. 1092.
[51] WT Grubb and CJ Michalske. “A high performance propane fuel cell operating in the
temperature range of 150–200 C”. In: Journal of The Electrochemical Society 111.9
(1964), p. 1015.
[52] Elton J Cairns. “Hydrocarbon fuel cells with fluoride electrolytes”. In: Journal of The
Electrochemical Society 113.11 (1966), p. 1200.
111
[53] FJ Rodr´ıguez Varela and O Savadogo. “The effect of anode catalysts on the behavior
of low temperature direct propane polymer electrolyte fuel cells (DPFC)”. In: Journal
of New Materials for Electrochemical Systems 9.2 (2006), p. 127.
[54] WS Li et al. “Chemicals and energy co-generation from direct hydrocarbons/oxygen proton exchange membrane fuel cell”. In: Journal of power sources 145.2 (2005),
pp. 376–382.
[55] Shadi Vafaeyan, Alain St-Amant, Marten Ternan, et al. “Nickel alloy catalysts for the
anode of a high temperature PEM direct propane fuel cell”. In: Journal of Chemistry
2014 (2014).
[56] H Khakdaman, Y Bourgault, and M Ternan. “Computational modeling of a direct
propane fuel cell”. In: Journal of Power Sources 196.6 (2011), pp. 3186–3194.
[57] Ermete Antolini. “Direct propane fuel cells”. In: Fuel 315 (2022), p. 123152.
[58] Eugene H. Kong. “Dynamics Of Direct Hydrocarbon Polymer Electrolyte Membrane
Fuel Cells”. PhD thesis. University of Southern California, 2019.
[59] Eugene H. Kong et al. “Dynamics of direct hydrocarbon PEM fuel cells”. In: Applied
Energy [In Submission] (2024).
[60] Scribner LLC. 850 fuel cell test system. 2023. url: https://www.scribner.com/
products/fuel-cell-test-systems/850-fuel-cell-test-system/.
[61] Rick Bitter, Taqi Mohiuddin, and Matt Nawrocki. LabVIEW: Advanced programming
techniques. Crc Press, 2006.
[62] url: https : / / www . fuelcellstore . com / fuel - cell - components / membrane -
electrode-assembly/hydrogen-air-oxygen-fuel-cell-mea/hydrogen-air-mea.
[63] Milton Stern and Al L Geary. “Electrochemical polarization: I. A theoretical analysis
of the shape of polarization curves”. In: Journal of the electrochemical society 104.1
(1957), p. 56.
[64] Balaji Krishnamurthy and S Deepalochani. “Performance of platinum black and supported platinum catalysts in a direct methanol fuel cell”. In: Int. J. Electrochem. Sci
4.3 (2009), p. 386.
[65] Z Belkhiri et al. “Effect of temperature and water content on the performance of PEM
fuel cell”. In: Journal of Renewable Energies 14.1 (2011), pp. 121–130.
[66] Lixin Fan, Zhengkai Tu, and Siew Hwa Chan. “Recent development of hydrogen and
fuel cell technologies: A review”. In: Energy Reports 7 (2021), pp. 8421–8446.
112
[67] Wenshang Chen et al. “Experimental study on dynamic response characteristics and
performance degradation mechanism of hydrogen-oxygen PEMFC during loading”. In:
International Journal of Hydrogen Energy 48.12 (2023), pp. 4800–4811.
[68] WRW Daud et al. “PEM fuel cell system control: A review”. In: Renewable Energy
113 (2017), pp. 620–638.
[69] J Hamelin et al. “Dynamic behavior of a PEM fuel cell stack for stationary applications”. In: International Journal of Hydrogen Energy 26.6 (2001), pp. 625–629.
[70] Claire H Woo and Jay B Benziger. “PEM fuel cell current regulation by fuel feed
control”. In: Chemical engineering science 62.4 (2007), pp. 957–968.
[71] Li Deng et al. “Dynamic modeling of a PEM fuel cell stack thermal system”. In: 2010
2nd International Conference on Advanced Computer Control. Vol. 2. IEEE. 2010,
pp. 473–477.
[72] JJ Baschuk and Xianguo Li. “Modelling of polymer electrolyte membrane fuel cells
with variable degrees of water flooding”. In: Journal of power sources 86.1-2 (2000),
pp. 181–196.
[73] JO Schumacher et al. “Control of miniature proton exchange membrane fuel cells based
on fuzzy logic”. In: Journal of Power Sources 129.2 (2004), pp. 143–151.
[74] Federico Zenith and Sigurd Skogestad. “Control of fuel cell power output”. In: Journal
of Process Control 17.4 (2007), pp. 333–347.
[75] Pauli Virtanen et al. “SciPy 1.0: Fundamental Algorithms for Scientific Computing in
Python”. In: Nature Methods 17 (2020), pp. 261–272. doi: 10.1038/s41592- 019-
0686-2.
[76] Simulink Documentation. Simulation and Model-Based Design. 2020. url: https :
//www.mathworks.com/products/simulink.html.
[77] Max Schwenzer et al. “Review on model predictive control: An engineering perspective”. In: The International Journal of Advanced Manufacturing Technology 117.5
(2021), pp. 1327–1349.
[78] Jean-Pierre Sleiman et al. “A unified mpc framework for whole-body dynamic locomotion and manipulation”. In: IEEE Robotics and Automation Letters 6.3 (2021),
pp. 4688–4695.
[79] Joel A E Andersson et al. “CasADi – A software framework for nonlinear optimization
and optimal control”. In: Mathematical Programming Computation 11.1 (2019), pp. 1–
36. doi: 10.1007/s12532-018-0139-4.
113
[80] Saptarshi Basu et al. “In situ optical diagnostics for measurements of water vapor
partial pressure in a PEM fuel cell”. In: (2006).
114
Abstract (if available)
Abstract
This dissertation investigates using chemical fuels, namely hydrogen and propane, for small scale power generation because of their high specific energy. It is focused on two projects: one using hydrogen with catalytic combustion to power microrobots, and another using propane in proton exchange membrane fuel cells (PEMFC) for portable power.
Most current microrobotic systems cannot carry enough onboard power, so they rely on stationary energy sources. This study presents a new 7-mg SMA-based miniature catalytic-combustion engine for millimeter-scale robotic actuation. This engine is made of a looped NiTi-Pt composite wire and a flat carbon-fiber beam, acting as a leaf spring. It can lift 650 times its own weight, producing 39.5 µW of average power. The design and performance of this engine are supported by a detailed model of heat transfer processes and SMA wire dynamics.
The work also explores low-temperature direct hydrocarbon PEMFCs as a feasible option for portable power generation. These cells are easier to store and handle than hydrogen PEMFCs. Despite their potential, issues like low power densities and the risk of the cell stopping power production exist. A custom experimental setup was created to study the dynamics and control of direct propane PEMFCs. The results offer new insights, differing from previous studies, and suggest ways to improve power output through control strategies.
In summary, the study provides new solutions, analytical models, and experimental setups to address challenges in microrobotics and portable power generation, focusing on the practical application of chemical fuels.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Dynamics of direct hydrocarbon polymer electrolyte membrane fuel cells
PDF
Mesoscale SOFC-based power generator system: modeling and experiments
PDF
Development of biologically-inspired sub-gram insect-scale autonomous robots
PDF
Design of nanomaterials for electrochemical applications in fuel cells and beyond
PDF
Studies of Swiss-roll combustors for incineration and reforming applications
PDF
Development of polystyrene sulfonic acid-polyvinylidene fluoride (PSSA-PVDF) blends for direct methanol fuel cells (DMFCS)
PDF
Studies on direct oxidation formic acid fuel cells: advantages, limitations and potential
PDF
Studies on direct methanol, formic acid and related fuel cells in conjunction with the electrochemical reduction of carbon dioxide
PDF
Studies on the flame dynamics and kinetics of alcohols and liquid hydrocarbon fuels
PDF
Studies of transport phenomena in hydrotalcite membranes, and their use in direct methanol fuel cells
PDF
Studies of siloxane decomposition in biomethane combustion
PDF
Electrocatalysts for direct liquid-feed fuel cells and advanced electrolytes for lithium-ion batteries
PDF
Design and modification of electrocatalysts for use in fuel cells and CO₂ reduction
PDF
Electron transfer capability and metabolic processes of the genus Shewanella with applications to the optimization of microbial fuel cells
PDF
Accuracy and feasibility of combustion studies under engine relevant conditions
PDF
Flame ignition studies of conventional and alternative jet fuels and surrogate components
PDF
The modification of catalysts and their supports for use in various fuel cells
PDF
Integrated carbon dioxide capture and utilization: catalysis enabled carbon-neutral methanol synthesis and hydrogen generation
PDF
Flame characteristics in quasi-2D channels: stability, rates and scaling
PDF
Electrokinetic transport of Cr(VI) and integration with zero-valent iron nanoparticle and microbial fuel cell technologies for aquifer remediation
Asset Metadata
Creator
Maimani, Fares
(author)
Core Title
Design, dynamics, and control of miniature catalytic combustion engines and direct propane PEM fuel cells
School
Viterbi School of Engineering
Degree
Doctor of Philosophy
Degree Program
Mechanical Engineering
Degree Conferral Date
2024-05
Publication Date
04/12/2024
Defense Date
04/05/2024
Publisher
Los Angeles, California
(original),
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
catalytic combustion,direct hydrocarbon proton exchange membrane (PEM) fuel cells,direct propane proton exchange membrane (PEM) fuel cells,dynamics and controls hydrocarbon PEM fuel cells,micro robotics,miniature combustion engine,OAI-PMH Harvest,portable power
Format
theses
(aat)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Ronney, Paul (
committee chair
), Prakash, Surya (
committee member
), Sadhal, Satwindar (
committee member
)
Creator Email
fares.maimani@gmail.com,maimani@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-oUC113871459
Unique identifier
UC113871459
Identifier
etd-MaimaniFar-12800.pdf (filename)
Legacy Identifier
etd-MaimaniFar-12800
Document Type
Dissertation
Format
theses (aat)
Rights
Maimani, Fares
Internet Media Type
application/pdf
Type
texts
Source
20240412-usctheses-batch-1139
(batch),
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Repository Email
cisadmin@lib.usc.edu
Tags
catalytic combustion
direct hydrocarbon proton exchange membrane (PEM) fuel cells
direct propane proton exchange membrane (PEM) fuel cells
dynamics and controls hydrocarbon PEM fuel cells
micro robotics
miniature combustion engine
portable power