Development and Applications of Chemical Sensors for the Detection of
Atmospheric Carbon Dioxide and Methane
By
Wesley T. Honeycutt
Bachelor of Science in Chemistry
University of Oklahoma
Norman, OK
2011
Submitted to the Faculty of the
Graduate College of the
Oklahoma State University
in partial fulfillment of
the requirements for
the Degree of
Doctor of Philosophy
May 2017
Development and Applications of Chemical Sensors for the Detection of
Atmospheric Carbon Dioxide and Methane
Dissertation Approved:
Dr. Nicholas F. Materer
Dissertation Adviser
Dr. Allen Apblett
Dr. Christopher J. Fennell
Dr. Jeffrey L. White
Dr. M. Tyler Ley
ii
ACKNOWLEDGEMENTS
The author would like to acknowledge a few people who have been instrumental to this work.
Notably, my advisor, Dr. Materer, has been essential to getting this project to move forward
and be successful. I would also like to acknowledge my committee members. The author
would also like to thank support members of various groups which have participated on this
project including: Dr. Peter Clark-project PI, Leonard Garcia-site liason for field work in
Farnsworth, TX, Dr. Jamey Jacob-site liason for field work at UAV airfield and co-PI, Jake
LeFlore-research specialist for Civil Engineering, Taehwan Kim-postdoc on this project with
Dr. Ley, Pouya Amrollahi-research assistant on this project with Dr. Ley, Xiaodan ’Sonia’
Li-postdoc with Dr. Ley who provided data analysis, Taylor Mitchell-Aerospace engineer
working on this project, David Porter-lab manager for Civil Engineering, Larry Vaughn, and
all members of the Physical Sciences instrument shop. The author is thankful for support
from various individuals outside of this project who have contributed to it in some way
including Dr. Claire M. Curry, Janet Honeycutt, Shoaib Shaikh, lab-mates past and present,
fellow Chemistry Department graduate students, and Leon.
Acknowledgements reflect the views of the author and are not endorsed by committee
members or Oklahoma State University.
iii
Name: Wesley T. Honeycutt
Date of Degree: May 2017
Title of Study: Development and Applications of Chemical Sensors for the
Detection of Atmospheric Carbon Dioxide and Methane
Major Field: Chemistry
Abstract:
This is a description of the design of a low-power, low-cost networked array of sensors for
the remote monitoring of carbon dioxide and methane. The goal was to create a scalable
self-powered two-dimensional array for the detection of these gases in a large area. The
sensor selection, electronic design, and data communication was studied and optimized
to allow for multiple units to form a self-assembling network for acre-scale coverage with
minimal human intervention. The final electronic design of the solar-powered units is flexible,
providing a foundation for future field deployable remote monitoring devices. Sensors were
selected for this application from commercially available models based on low-power, low-
cost, market availability, detection range, and accuracy around the global baseline criteria.
For environmental monitoring, carbon dioxide sensors are characterized near 400 ppm and
methane from 2 to 200 ppm. For both gases, exertions up to several 1000 pm were examined
to mimic large releases. An Xbee mesh network of radios was utilized to coordinate the
individual units in the array, and the data was transferred in real-time over the cellular
network to a dedicated server. The system was tested at a site north of the Oklahoma State
campus, an unmanned airfield east of Stillwater, OK, and an injection well near Farnsworth,
TX. Data collected from the Stillwater test sites show that the system is reliable for baseline
gas levels. The gas injection well site was monitored as a potential source of carbon dioxide
and methane leaks due to the carbon dioxide injection process undertaken there for carbon
sequestration and enhanced oil recovery efforts. The sensors are shown to be effective at
detecting gas concentration at the sites and few possible leak events are detected.
iv
TABLE OF CONTENTS
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Subterranean Carbon Sequestration . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Monitoring Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Comparison of Commercially Available CO
2
and CH
4
sensors . . . . . . . . . . . 10
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3.1 Precision and Baseline Noise Tests . . . . . . . . . . . . . . . . . . . 16
2.3.2 Limit of Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.1 Precision and Baseline Noise Tests . . . . . . . . . . . . . . . . . . . 21
2.4.2 Response Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4.3 Limit of Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3
Determining the Battery Shutoff Circuit Behavior of a Commercially Available Solar
Charging Circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2 Discharge Behavior and Cutoff Point . . . . . . . . . . . . . . . . . . . . . . 32
3.3 A Direct Comparison of Batteries and Damage Due to Repeated Discharge . 32
3.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4 Networking with from an Xbee Modem Mesh to a Cellular Modem . . . . . . . . 38
4.1 Network Topologies for Remote Sensing . . . . . . . . . . . . . . . . . . . . . 38
4.2 Modems and Radios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.3 Data Storage and Transmission . . . . . . . . . . . . . . . . . . . . . . . . . 46
5
Development of a Networked Gas Sensor Array for the Detection of CO
2
and CH
4
Concentrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.2 Device Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
v
5.2.1 Gas Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.2.2 Power Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.3 Circuit Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.3.1 Power Supply . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.3.2 Processor and Memory . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.3.3 XBee Wireless Communication . . . . . . . . . . . . . . . . . . . . . 60
5.3.4 Cellular Communication . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.3.5 Sensor Breakout Board . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.3.6 Gascard Implementation . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.3.7 Ground Protection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.4 Air Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.4.1 Passive Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.4.2 3D Printed Parts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.4.3 Active Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.5 Prototyping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.6 Mass Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6
Results of Prototyping Tests and Long-term Tests of a Networked Sensor Array at
a Proving Ground on the OSU Campus . . . . . . . . . . . . . . . . . . . . . . . . 93
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.2 Preliminary Deployment and Unscheduled Mechanical Stress Testing . . . . 93
6.2.1 Detected Events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.2.2 Storm Damage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.3
Comparing Collected Data from Long-Term Array against Accepted Weather
Report Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7
Results of Data Collected Tests from a Networked Sensor Array at the OSU
Unmanned Airfield . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
7.2 Sensor Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.3 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.3.1 Carbon Dioxide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.3.2 Methane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.3.3 Peak Events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
7.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
8
Results of Data Collected from a a Networked Sensor Array near a Gas Injection
Well Field Site near Farnsworth, TX . . . . . . . . . . . . . . . . . . . . . . . . . 118
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
8.2 Sensor Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
8.3 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
vi
8.3.1 Carbon Dioxide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
8.3.2 Methane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
8.3.3 Peak Events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
8.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
9.2 Summary of Work and Findings . . . . . . . . . . . . . . . . . . . . . . . . . 129
9.3 Beyond the Project Goals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
10 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
A Sensor Board - Primary Circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
B Sensor Board - Sensors Breakout Circuit . . . . . . . . . . . . . . . . . . . . . . . 161
C Sensor Board - Cellular Modem Breakout Circuit . . . . . . . . . . . . . . . . . . 166
D OpenSCAD Code Used for 3D Printed Part . . . . . . . . . . . . . . . . . . . . . 167
E Assembly Instructions For Sensor Nodes . . . . . . . . . . . . . . . . . . . . . . . 172
VITA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
vii
List of Figures
1.1 A summary of carbon dioxide injection wells . . . . . . . . . . . . . . . . 3
1.2 The steps in subterranean plume migration . . . . . . . . . . . . . . . . 5
2.1 Gas Mixing Apparatus Diagram . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Carbon dioxide baselines . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Methane baselines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4 Carbon dioxide Gaussian distribution . . . . . . . . . . . . . . . . . . . . 19
2.5 Methane Gaussian distriution . . . . . . . . . . . . . . . . . . . . . . . . 27
2.6 Gas Mixing Apparatus . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.7 Trendline comparision of sensor responses . . . . . . . . . . . . . . . . . 29
2.8 MQ-4 concentration step plot showing overshoot . . . . . . . . . . . . . 30
3.1 Battery shutoff behavior by protection circuit . . . . . . . . . . . . . . . 33
3.2 Shutoff behavior of battery circuit with a 141 Ω load . . . . . . . . . . . 34
3.3 Shutoff behavior of battery circuit with a 188 Ω load . . . . . . . . . . . 35
3.4 Shutoff behavior of battery circuit with a 235 Ω load . . . . . . . . . . . 35
3.5 Battery discharge time vs. current . . . . . . . . . . . . . . . . . . . . . 36
3.6 Battery degredation upon repeated tests . . . . . . . . . . . . . . . . . . 37
4.1 Point-to-point Network Topology . . . . . . . . . . . . . . . . . . . . . . 39
4.2 Line Network Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.3 Bus Network Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.4 Star Network Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.5 Mesh Network Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.6 Hybrid Network Topology . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.7 Xbee modem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
viii
4.8 Xbee Mesh Sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.9 Sensor and Communication Nodes Functional Diagram . . . . . . . . . . 46
4.10 Data Collection Scheme - Sensor Node . . . . . . . . . . . . . . . . . . . 47
4.11 Data Collection Scheme - Communication Node . . . . . . . . . . . . . . 48
4.12 Data Collection Scheme - Server Storage . . . . . . . . . . . . . . . . . . 49
5.1 Equipment outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.2 Tycon Power Systems enclosures . . . . . . . . . . . . . . . . . . . . . . 54
5.3 Control board . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.4 Control board - Sensor node parts . . . . . . . . . . . . . . . . . . . . . 57
5.5 Control board - Communicaiton node parts . . . . . . . . . . . . . . . . 57
5.6 Circuit - Fuse protected 12 V terminal circuit . . . . . . . . . . . . . . . 58
5.7 Circuit - 3.3 V DC/DC converter circuit . . . . . . . . . . . . . . . . . . 58
5.8 Circuit - 5 V Dc/DC converter circuit . . . . . . . . . . . . . . . . . . . 60
5.9 Circuit - External crystal oscillator . . . . . . . . . . . . . . . . . . . . . 60
5.10 Circuit - Example of switch and associated LED . . . . . . . . . . . . . 61
5.11 Circuit - ICSP header pinout . . . . . . . . . . . . . . . . . . . . . . . . 61
5.12 Circuit - DS3231 real time clock circuit . . . . . . . . . . . . . . . . . . . 61
5.13 Circuit - EEprom memory IC and connections . . . . . . . . . . . . . . . 62
5.14 Circuit - microSD memory slot with switch circuitry . . . . . . . . . . . 62
5.15 Cellular modem breakout board . . . . . . . . . . . . . . . . . . . . . . . 63
5.16 Cellular modem part . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.17 Circuit - Schematic of XBee device . . . . . . . . . . . . . . . . . . . . . 64
5.18 Circuit - Cellular communication breakout board schematic . . . . . . . 65
5.19 Sensor breakout board . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.20 K-30 Carbon dioxide part . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.21 Circuit - Carbon dioxide sensor with level translator . . . . . . . . . . . 67
5.22 Circuit - Mainboard carbon dioxide level translator . . . . . . . . . . . . 67
5.23 Mq-4 methane sensor part . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.24 Mq-4 methane sensor part with sensing element exposed . . . . . . . . . 68
5.25 Circuit - methane sensor socket with amplifier . . . . . . . . . . . . . . . 69
5.26 Circuit - SPI and level translator for methane sensor . . . . . . . . . . . 69
ix
5.27 Sensiriron SHT75 temperature sensor part . . . . . . . . . . . . . . . . . 70
5.28 Freescale MPXS6115 pressure sensor part . . . . . . . . . . . . . . . . . 70
5.29 Circuit - SHT75 schematic on sensor board . . . . . . . . . . . . . . . . 71
5.30 Circuit - MOSFET switches used with the SHT75 clock line . . . . . . . 71
5.31 Circuit - SHT75 data level translator on mainboard . . . . . . . . . . . . 71
5.32 Circuit - MPXA6115A pressure sensor and amplifier . . . . . . . . . . . 72
5.33 Circuit - Gascard 4 V power supply . . . . . . . . . . . . . . . . . . . . . 72
5.34 Circuit - Pump power and adjustment . . . . . . . . . . . . . . . . . . . 73
5.35 Circuit - Serial to asynchronous communication translator . . . . . . . . 74
5.36 Labeled cartoon of 3D printed part . . . . . . . . . . . . . . . . . . . . . 77
5.37 Early 3D printed part produciton . . . . . . . . . . . . . . . . . . . . . . 78
5.38 Closeup of Makerbot 2 detail . . . . . . . . . . . . . . . . . . . . . . . . 79
5.39 Closeup of Makerbot 2 detail, curved . . . . . . . . . . . . . . . . . . . . 80
5.40 Closeup of EOSINT detail . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.41 Closeup of EOSINT detail, curved . . . . . . . . . . . . . . . . . . . . . 82
5.42 Early prototype of plastic parts . . . . . . . . . . . . . . . . . . . . . . . 82
5.43 3D printed part attached to the board . . . . . . . . . . . . . . . . . . . 83
5.44 3D printed part with silicone . . . . . . . . . . . . . . . . . . . . . . . . 83
5.45 3D printed part joins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.46 Lexan box housing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.47 Arduino Mega development board . . . . . . . . . . . . . . . . . . . . . 86
5.48 Internals of sensor node . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.49 A delivery of control boards from the assemblers . . . . . . . . . . . . . 89
6.1 Long term field testing site . . . . . . . . . . . . . . . . . . . . . . . . . 94
6.2 Carbon dioxide detection by a prototype units . . . . . . . . . . . . . . . 95
6.3 Methane detection by two prototype units . . . . . . . . . . . . . . . . . 96
6.4 Storm damage to prototype sensors . . . . . . . . . . . . . . . . . . . . . 97
6.5 Storm damage to UNIT 0003 - 1 . . . . . . . . . . . . . . . . . . . . . . 98
6.6 Storm damage to UNIT 0003 - 2 . . . . . . . . . . . . . . . . . . . . . . 99
6.7 Storm damage to UNIT 0004 . . . . . . . . . . . . . . . . . . . . . . . . 100
6.8 Temperature data collected vs. weather data . . . . . . . . . . . . . . . 102
x
6.9 Humidity data collected vs. weather data . . . . . . . . . . . . . . . . . 103
6.10 Pressure data collected vs. weather data . . . . . . . . . . . . . . . . . . 104
6.11 Carbon Dioxide Concentration at Proving Ground . . . . . . . . . . . . 105
6.12 Methane Concentration at Proving Ground . . . . . . . . . . . . . . . . 106
7.1 UAV Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.2 Map of Airport Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
7.3 Distribution of Sensor Responses in a Daily Cycle . . . . . . . . . . . . . 111
7.4 Aligned Data from UAS CO
2
Sensors . . . . . . . . . . . . . . . . . . . . 112
7.5 Averaged Response from all UAS CO
2
Sensors . . . . . . . . . . . . . . 113
7.6 Distribution of Responses from UAS CO
2
Sensors . . . . . . . . . . . . . 113
7.7 Aligned Data from UAS CH
4
Sensors . . . . . . . . . . . . . . . . . . . . 114
7.8 Averaged Response from all UAS CH
4
Sensors . . . . . . . . . . . . . . . 115
7.9 Distribution of Responses from UAS CH
4
Sensors . . . . . . . . . . . . . 115
7.10 UAS Detected Events . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
8.1 Proposed grid of sensors at Farnsworth, TX site . . . . . . . . . . . . . . 119
8.2 Final sensor placement at Farnsworth, TX site . . . . . . . . . . . . . . 120
8.3 Mounting of Sensors on Power Poles . . . . . . . . . . . . . . . . . . . . 121
8.4 Distribution of Sensor Responses in a Daily Cycle . . . . . . . . . . . . . 123
8.5 Aligned Data from Farnsworth, TX CO
2
Sensors . . . . . . . . . . . . . 123
8.6 Averaged Response from all Farnsworth, TX CO
2
Sensors . . . . . . . . 124
8.7 Distribution of Responses from Farnsworth, TX CO
2
Sensors . . . . . . 125
8.8 Aligned Data from Farnsworth, TX CH
4
Sensors . . . . . . . . . . . . . 125
8.9 Averaged Response from all Farnsworth, TX CH
4
Sensors . . . . . . . . 126
8.10 Distribution of Responses from Farnsworth, TX CH
4
Sensors . . . . . . 127
8.11 Farnsworth, TX Detected Events . . . . . . . . . . . . . . . . . . . . . . 128
E.1 Instructions for sensor node assembly, slides 1-6. . . . . . . . . . . . . . 173
E.2 Instructions for sensor node assembly, slides 7-12. . . . . . . . . . . . . . 174
E.3 Instructions for sensor node assembly, slides 13-18. . . . . . . . . . . . . 175
E.4 Instructions for sensor node assembly, slides 19-24. . . . . . . . . . . . . 176
E.5 Instructions for sensor node assembly, slides 25-30. . . . . . . . . . . . . 177
xi
E.6 Instructions for sensor node assembly, slides 31-36. . . . . . . . . . . . . 178
E.7 Instructions for sensor node assembly, slides 37-42. . . . . . . . . . . . . 179
E.8 Instructions for sensor node assembly, slides 43-48. . . . . . . . . . . . . 180
E.9 Instructions for sensor node assembly, slides 49-54. . . . . . . . . . . . . 181
E.10 Instructions for sensor node assembly, slides 55-60. . . . . . . . . . . . . 182
xii
List of Tables
2.1 Manufacturer listed properties of evaluated carbon dioxide sensors . . . . 13
2.2 Manufacturer listed properties of evaluated methane sensors . . . . . . . 13
2.3 Ratios of calibrated gases used in mixed gas experiments. . . . . . . . . . 14
2.4 Precision and Accuracy of Carbon Dioxide Sensors . . . . . . . . . . . . . 23
2.5 Precision and Accuracy of Methane Sensors . . . . . . . . . . . . . . . . . 23
2.6 Detection Limits of Carbon Dioxide Sensors . . . . . . . . . . . . . . . . 25
2.7 Detection Limits of Methane Sensors . . . . . . . . . . . . . . . . . . . . 25
5.1 Device hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.2 Power use of sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
xiii
Chapter 1
Introduction
Great strides are made in science which provide Man with more capabilities to measure
the environment. To make best use of these advances, a determined approach must be made
to the design of systems applying these tools. The design of a proper system using advanced
sensing techniques is essential to ensuring that the core science is able to benefit the users.
The work reported in this dissertation is a summary of projects related to the application of
existing gas sensors to the detection of microseepage of carbon dioxide and methane from
subterranean storage sites. This work will discuss comparison of several individual gas sensors,
the design of a networked array which utilizes these sensors, construction of the individual
node units of this network, deployment of the networks at several sites, and the data resulting
from experiments at these sites.
1.1 Subterranean Carbon Sequestration
The effect of carbon dioxide in the atmosphere on terrestrial temperature has been
known to science since the late 19
th
century as quantified by Arrhenius [
1
]. As carbon
dioxide output from anthropogenic sources increases [
2
], it becomes increasingly important to
find solutions to prevent excessive buildup. One such proposed solution is injection of excess
carbon dioxide underground. There, the injected gas can be stored in naturally occurring
subterranean voids or those left by mining and hydrocarbon extraction processes. Alternately,
the carbon dioxide can be used as an extraction fluid to pull valuable hydrocarbons such as
methane still remaining in these wells in a process termed Enhanced Oil Recovery (EOR).
1
Sites of particular interest are bituminous sands, unreachable coal beds, deep saline aquifers,
and salt caverns [3, 4].
In sequestration efforts, carbon dioxide is collected from significant waste streams such
as those produced by coal-fired power plants, and injected into underground voids [
4
]. The
procedure involves pressurization of carbon dioxide into a supercritical fluid and injection
into a drilled hole. As the fluid is pumped into the well, it expands to fill porous regions
as the pressure reaches a penetration threshold. It is theorized that the majority of the
injected gas will be converted from fluid carbon dioxide to carbonate salts with time after
the injection. By this process, the waste streams from large industrial producers of carbon
dioxide is converted from an emitted greenhouse gas pollutant to a more manageable solid
form. As it is accepted that the injected carbon dioxide will remain underground for many
years, it is considered to be a viable solution to excess emissions. Studies comparing the
collection and injection methods of the gas into different storage locations at small scale
suggest that carbon sequestration by the injection of flue gas into underground voids is a
net benefit in terms of greenhouse gas mediation efforts [
5
]. A summary of this process is
depicted in Figure 1.1. In a secondary process, as the carbon dioxide seeps from the storage
depth, it undergoes biogenic reduction to methane, an even more potent greenhouse gas [
6
].
Most of this gas is expected to remain in the storage site as well, but reliable sensing method
for detecting problems due to changes in flux of both these gases is essential to the safe
deployment of this method in the interest of environmental health and safety.
The pressurized carbon dioxide gas can be used to liberate hydrocarbons from the
ground by treating the gas as an extraction solvent rather than a permanent injectable. This
is the premise of the use of carbon dioxide injection as a method of enhanced oil recovery
(EOR). Pressurized carbon dioxide is pumped into new or existing wells and the fluid fills in
spaces in the surrounding earth, replacing an existing material which is easily dissolved into
the carbon dioxide [
7
] and potentially liberating valuable gases trapped in these underground
areas such as methane [
8
]. Additionally, a variant of this process can be used to extract
crude oil which has not been liberated by traditional drilling. Most EOR projects occur
at existing oil and gas developments or at “depleted” sites [
9
]. A report from the National
Energy Technology Laboratory of the United States Department of Energy claims that the
sequestration of 20 billion metric tons of carbon dioxide has been employed as part of EOR
2
Figure 1.1: This diagram shows a simplified explanation of the carbon dioxide injection process. The
cartoon depicts carbon dioxide being collect from a large industrial source, e.g. a power plant, and pumped
into a deep saline aquifer well. The gas is injected as a supercritical fluid near the top of a void surrounded
by a sturdy rock formation. The pressure from the expanding fluid pressurizes the well. Remaining oil in
the well sitting in a layer above the aquifer is easily pumped out by existing infrastructure.
programs since their discovery [
10
,
11
]. As of 2014, 53% of all commercial-scale EOR projects
in the United States utilize gas injection techniques, and the number of projects using this
method is expected to increase as there is more push for carbon dioxide sequestration in
conjunction with rising energy prices [
12
]. Unlike environmental remediation sequestration
efforts, the majority of the carbon dioxide gas for EOR projects is pumped from geologic
sources. Only a third of the gas used in small-scale EOR research wells was procured from
flue gas [9].
There are some concerns regarding the potential outcomes of such storage regimes [
13
].
A large leak of the sequestered carbon dioxide which has not undergone conversion to carbonate
3
salts has the potential for catastrophic and expensive consequences. A plume of carbon
dioxide has the potential of suffocating all animal life in an area, as evidenced by the events at
Lake Nyos [
14
] and Lake Monoun [
15
] in Cameroon which caused massive loss of human life.
It is essential that this outcome be avoided, therefore it is equally essential to monitor the gas
flux from the storage sites over time. Small leaks over time, while less catastrophic in nature,
would undermine the assumption that carbon sequestration as a fail-proof storage mechanism
for excess carbon dioxide waste gas as the presumed net gain in environmental health would
be lost over longer periods. There have been no studies of carbon dioxide leakage from carbon
sequestration sites over long time scales. This is due, in part, to the recent development
of the technique. However, there are currently inadequate monitoring technologies for this
kind of study. Often, atmospheric ramifications of sequestration and EOR is ignored entirely.
Only half of small-scale EOR testing projects conducted atmospheric monitoring, instead
most rely on subsurface monitoring and geochemical modeling techniques [9].
If the fluidized carbon dioxide in the rocky subsurface environment can be thought
of as a mobile phase in a large solid substrate, it is reasonable to suspect that voids in the
solid material would physically separate as layers. It is proposed by some researchers that
the injected carbon dioxide will migrate towards the surface, eventually seeping out in small
quantities, in a process known as microseepage [
16
,
17
,
18
]. This is the vertical migration
of an underground “plume” which moves toward the surface. This plumed gas eventually
will leak from the subterranean storage site from small fissures in the ground. A simplified
cartoon of plume migration is depicted in Figure 1.2.
The migration of carbon dioxide is recognized as an important factor for determining
the economic feasibility of an injection well, and as a prospective legislative issue [
19
]. It is
difficult and costly to monitor these gas plumes experimentally. Instead, researchers rely
heavily on computer modeling, which has been shown to be effective for predicting the lateral
diffusion of the plume underground [
20
,
21
]. The scope of many theoretical studies of carbon
dioxide injection primarily focuses on lateral plume migration through individual layers,
assuming that there is minimal vertical migration [
22
]. While this is predominantly the case,
moderately porous materials with interconnected channels induce gas mobility to a similar
degree as inter-particulate spaces in laboratory testing [
23
]. Certain models for the prediction
of plume migration after injection are simplified by assumptions which may not accurately
4
Figure 1.2: This cartoon depicts a simplified illustration of the plume which forms during CO
2
injection.
In the leftmost image, a quantity of CO
2
is pumped in from the surface (represented somewhat inaccurately
by the oil derrick). As more CO
2
is left in the well, the supercritical fluid expands through crevices in the
rock and sediment layers predominantly toward the surface. Eventually, this stored gas will reach the surface
and begin to gradually seep out through a process known as microseepage.
model the physical world, including the assumption that the surface layer is impermeable to
the fluidized gas [
24
]. Fluidized carbon dioxide does not increase porosity of solid material,
however the focus on lateral plume migration in theoretical models does not adequately
address the potential for microseepage from subterranean storage environments.
1.2 Monitoring Strategies
Monitoring of gas flux transitioning from subsurface storage areas to the atmosphere
can take place using a variety of techniques at various locations around a wellhead and plume
site. Atmospheric monitoring methods include optical sensors such as non-dispersive infrared
(NDIR), light detection and ranging (LIDAR), sorbent sampling of tracer molecules, and
eddy covariance (EC) flux monitoring; near-surface monitoring methods include soil sampling,
flux accumulation, hyperspectral imaging, and many more [
25
]. While this list is limited
in scope, it touches upon some of the most commonly deployed strategies. Yet, there are
many shortcomings with the available technology. There is currently no option which allows
5
for sensitive detection of analyte gases at a short time-scale with remote monitoring for
long measuring periods. The current implementation of these strategies ranges from being
ineffective to outright hazardous.
The relative inaccessibility of many injection well sites poses a problem. Soil sampling
has long been used for determining the effects of wells on the local soil environment. Operators
go to field sites and take either core samples or headspace samples from multiple sites in
the area. While monitoring agencies and injection well operators can send employees to the
field sites to collect data, this is difficult and inefficient. There is a recent push to introduce
more remote monitoring technologies to reduce these shortcomings. Spectroscopy remains
an ever attractive options for remote sensing of local gas concentrations. Hyperspectral
imaging, LIDAR, and other similar technologies working in combination have been shown
to be powerful in detecting leak anomalies [
26
,
27
]. Yet these technologies often require a
trained operator to function. The recent availability of unmanned aerial systems has been of
particular interest for remote monitoring [
28
]. Yet these systems only allow for temporary
scanning of a single set of sites. Planes must be refueled, batteries recharged, and sensors
redeployed for each one of these scans. The attractiveness of unmanned aerial systems is
limited by the power constraints of these units. Continual monitoring of a site requires active
maintenance. Satellites in geosynchronous orbit allow for constant hovering above large areas.
A recent study by NASA of a methane leak has shown that the modern LIDAR equipment was
even able to measure an above-ground gas plume from a large leak [
29
]. The gas leak depicted
in the article by Thompson et al. was exceptionally large. Most leak events and microseepage
occurrences are likely very minor deviations from the local baseline. Current space based
spectrometers cannot detect near-baseline deviations of small leakage sites. Additionally, the
deploying a satellite is still prohibitively expensive. This means that continual monitoring of
every potential leakage site is bottlenecked by the available satellites. Custom built terrestrial
remote monitoring stations in development for the In Sallah injection well have been shown
to have strong detection capabilities in simulated leaks [
30
,
31
]. Yet these proposed sensors
are only capable of single site monitoring as of this publication. As potential leakage from
subterranean storage site may occur over a large area and from surrounding abandoned
wells [
32
,
33
], single site detection is not practical for monitoring of the large area potentially
affected by an injected carbon dioxide plume. Eddy Covariance is praised for the large area
6
that can be monitored, however the devices are maintenance intensive, requiring adequate
infrastructure in the sequestration area and computational power due to data complexity [
34
].
Current remote monitoring techniques may serve to reduce the man-hours required compared
to traditional soil sampling regimes, yet they fall short of the desired goals of true remote
detection of small local environmental changes.
Possibly the most selective method for detecting flux of injected chemicals to the
atmosphere is the use of tracer chemicals. Addition of a trace quantity of a chemical
which is inert to the system is made during the injection process. Sites can be monitored by
selective testing for these chemicals. The low signal-to-noise ratio of optical detection for tracer
chemicals in flux gas is an attractive option, but has the potential for environmental damage by
way the persistent pollutant nature of the perfluorocarbons used as tracers [
35
,
36
]. For crude
oil well-to-well analysis, it has long been established that the addition of an anthropogenic
radioisotope can be detected from pumped material based on characteristic radiation from the
sample. Only certain chemicals meet the needs of carbon dioxide injection wells [
37
]. Notable
among this subset is the use of halogenated organic compounds, sulfur hexafluoride, and
hydrogen sulfide. Many of these chemicals have been identified as environmentally hazardous
compounds. Many of the halogenated organic compounds recommended for use, often freon
variants, are persistent chemicals with high greenhouse gas potentials. Sulfur hexafluoride,
while attractive due to its inert qualities, is extremely persistent with high radiative forcing
such that it is considered to be more than 26,000 times more greenhouse warming potential
over the 100 years than carbon dioxide [
38
]. Some tracers, such as hydrogen sulfide, have
been shown to enter aquifers, causing them to be non-potable sources for future use [
39
].
These tracer chemicals may not even be effective in supercritical fluids. It is proposed by
Vanderweijer et al. that higher solubility of carbon dioxide in water than tracer chemicals
may lead to inaccurate concentration measurements in offshore injection environments and
wells known to contain significant water deposits [
19
]. An optimal solution to the monitoring
of injection wells would quantify the migrating gas without requiring an environmentally
persistent tracer.
There are no monitoring devices which can collect data from multiple sites simultane-
ously at fast time-scales over a continuous period. This is evidenced by the lack of available
data on the diel cycle of gas concentrations. Isolated studies have been conducted which have
7
established the day-night cycling of gases such as carbon dioxide and methane, yet these often
only measure a few points of data per 24-hour cycle. Determining the diel cycle for these
gases is valuable to climate scientists, biologists, and agricultural engineers, and data of the
diel cycle of concentration over various terrestrial and aquatic local environments is available
in the literature for carbon dioxide [
40
,
41
,
42
,
43
,
44
,
45
] and methane [
46
,
47
,
48
,
49
,
50
,
51
].
While only a sampling of the available literature is mentioned in the previous citations, a
critical analysis of all of these sources shows that current technologies are not utilized to
detect concentration often, for a long time, and over a large area. These studies usually
involve a single sampling instrument operated continuously for a short period, samples taken
with large gaps between points, or distributed detection of multiple sites not performed
concurrently. Recent developments of devices under the “internet of things” notion suggest
that construction of devices for this purpose should be simple. Yet no one has done it.
The specialized knowledge required to construct these devices, the programming experience
required to operate them, and the time required to perfect such devices to a field-deployable
state appears to have deterred many scientists from attempting to create units of this class.
The construction and characterization of instruments designed for distributed detection in
the scientific literature would provide a framework for other scientists-reducing the barrier to
these technologies for applications described in this monograph as well as further applications.
A technological niche exists for a relatively inexpensive device which can monitor a
large number of sites easily with little human interaction. A solution to this problem exists
in the application of available technology to develop a distributed sensing unit that wirelessly
transmits information to a gathering center. This idea in itself is not completely without
precedent. Wireless sensor arrays have been used before in agricultural applications to
monitor conditions [
52
] and aerial cloud analysis [
53
]. In these cases they have demonstrated
the capacity to perform required monitoring functions over a large area with low cost and
human interaction. An adaptation of this technology to fit with the needs of gas monitoring
at wellheads provides our solution.
To apply the wireless sensor array technology to environmental monitoring of gas
flux, we must consider the sensors to connect to the network. The sensors developed for this
project will directly measure the concentration of gas at the surface using a combination
of infrared sensors and atmospheric condition sensors. While the idea of deploying a large
8
number of NDIR sensors is not completely new either [
54
], the technology suggested for
these deployments (short closed-path NDIR) requires constant pumping to operate, making
it energetically expensive and impractical for a remote monitoring network. This project
will utilize a unique combination of the pumped closed path NDIR sensors and passive gas
sensors to reduce energy cost.
In this work, various sensors with purported detection capabilities near the global
baseline for carbon dioxide and methane will be investigated and directly compared in
Chapter 2. Sensors from this analysis will be used in the construction of the gas sensor
array. The low-power constraints of this array will be considered by testing batteries from a
commercially available solar-powered enclosure for shutoff thresholds and repeated discharge
damage in Chapter 3. The devices will be programmed for low-power operation based on
available battery charge from these data. The networking strategy for the proposed array
will be discussed and implemented in Chapter 4. A detailed description of the individual
circuitry of the devices will be presented in Chapter 5. This discussion of the device will also
include considerations made for sampling regimes and layout. The constructed devices in the
networked array of gas sensors will be deployed to three sites, a prototyping field north of
the OSU campus in Chapter 6, an unmanned aerial systems test field in Chapter 7, and a
carbon dioxide injection well near Farnsworth, TX in Chapter 8. At these sites, the data
produced by the sensors will be compared to known data from weather stations to confirm
that the arrays are producing results consistent with accepted data. The data produced by
sensors at the airfield and injection well sites will be characterized by statistical analysis and
event threshold determination.
In the appendices, information is made available to facilitate future replication of
the work described here. Complete schematics for the printed circuit boards used in these
devices are included in Appendix A, B, and C. A script used to generate a 3D printed part
for atmospheric interface is given in Appendix D. Detailed assembly instructions for the
devices used in this work are given in Appendix E.
9
Chapter 2
Comparison of Commercially Available CO
2
and CH
4
sensors
2.1 Introduction
Environmental monitoring of local gas concentration is becoming increasingly im-
portant to ensure worker safety and for early identification of potential leaks. Real time
monitoring of carbon dioxide and methane gas concentrations are of interest since these gases
impact animal health and crop growth. Since gases are considered greenhouse gases, monitor-
ing industrial sites are important in understand their impact on the environment. In designing
monitoring devices, the sensor for gas concentration analysis must be matched to the specific
operational requirements, such as precision, reliability, and power consumption. Example sen-
sor elements include those used in HVAC air handlers [
55
,
56
], chemical processing units[
57
],
oil well monitoring devices [
58
,
59
], and environmental monitoring [
60
,
61
,
62
,
63
,
64
]. Given
the myriad of other applications, the uses cited represent only a small sample of the various
applications and varieties of sensor elements. Despite the large number of sensors which are
commercially available, there is limited literature that directly compares sensors. This lack of
information limits ones ability to select the optimal sensor for a given application. To address
this issue, this paper compares an array of commercially available, low-cost sensors for carbon
dioxide (chemical formula CO
2
) and methane (chemical formula CH
4
) at concentration typical
to those required for environmental monitoring around carbon dioxide sequestering operations
and oil fields.
Potential sensors range from small, inexpensive chemiresistive sensors to complex
10
and costly optical systems. The low-cost chemiresistive based methane sensors are typically
used in gas warning systems [
65
]. On the other end, Light Imaging Detection and Ranging
(LIDAR) is an accurate and effective method for remote monitoring of industrial sites, for
example oil wells [
66
]. However, these devices are limited by cost and operational complexity,
and are not suitable for portable self-powered monitoring devices. The sensors need good
sensitivity and precision around the baseline atmospheric concentration for each analyte,
which is around 400 ppm for carbon dioxide [
67
,
68
] For methane, the baseline atmospheric
concentration is under 2 ppm [
69
,
70
,
71
]. Thus, for methane, sensors were selected and tested
near their baseline. In addition to concentration, power consumption, reliability, and ease of
integration are also important factors. Sensors will be selected based on their sensitivity at
atmospheric levels, ease of use, power consumption, price, and market availability.
2.2 Objectives
The objective of this study is to determine the suitability of commercially available
low-cost sensors for integration into portable self-powered monitoring equipment. The low-
cost sensors currently available can be generally divided by detection method, either optical
absorption or electrical changes due to a chemical reaction with the analyte (chemiresistive).
Previous researchers have cited concerns with electrochemical sensors for these gases, as they
have a short lifetime and lack robustness [
72
]. The selection process is restricted to sensors
that are commercially available in large volumes (at least 1000 units) at low-cost (defined
here as less than $100 per unit in bulk).
Optical sensors typically have excellent sensitivity, selectivity, and fast response to
concentration changes. Optical methods of detecting carbon dioxide and methane are based
on the measuring the absorption of light at
2352 cm
−1
and
3015 cm
−1
, respectively [
73
,
74
].
Sensors utilizing these specific wavelengths are termed dispersive infrared instruments and
are typically bulky, costly and fragile. Low-cost sensors utilize nondispersive infrared (NDIR)
sensing, which utilizes a source which emits at a broad spectrum coupled with a narrow band
pass filter across the absorbance maximum. As a result, lower cost parts can be used and the
design can be both more compact and robust. These optical methods utilize the Beer-Lambert
Law to relate absorptions to concentrations, and thus are only dependent on the geometry
of the sensor and physical properties of the gas. [
75
] In general, NDIR detection is utilized
11
for carbon dioxide do due its relatively large molar adsorption coefficient results in a short
path length. Methane is limited in practical applications due to its low absorbance coefficient
and overlap interfered symmetric C-H stretches, making methane difficult to distinguish
from ethane or propane, for example. [
76
]. Small chemiresistive sensors for carbon dioxide
and methane are primarily based on reactions between the analyte and a semiconductor
film, typically tin oxide. [
72
] Although, chemiresistive sensors have significant drawbacks,
including selectivity, these sensors are easy to produce and have very low cost, leading to
widespread use. Commercially, there are several available chemiresistive for detection of
methane. These sensors work by measuring changes in electron transport through the metal
oxide semi-conductive film, in the presence of oxygen and reactive gases [
77
]. When the film
is exposed to methane, the molecule adsorbs and reacts with surface oxygen species. This
liberates free electrons in the bulk and reduces the electrical resistance [78, 79].
To drive the chemical reactions at a reasonable rate, the film must be heated to at
least
350
◦
C
to work effectively [
80
]. drastically increases power consumption. Additionally,
the measured transduction depends on the surface of the film, the presence of oxygen
containing species, and the reaction rate. As such, the output depends on the relative
humidity, temperature, and preparation of the film. Unlike optical sensors which depend
only on physical properties of the analyte, the output of these sensors have a complex
relationship to the gas concentration which varies from sensor to sensor. For carbon dioxide,
the authors located a chemiresistive sensor which operates slightly different. This variant
detects resistance changes that occur when carbon dioxide adsorbs and reacts to form a
carbonate on the surface of the film [81].
2.3 Methods
To test the sensors under controlled conditions, an experimental setup (Figure 2.1)
was constructed. This setup allows a known concentration to be prepared from a calibrated
gas cylinder. A high-quality bench-top analyzer (California Analytical Instruments, Inc.
ZRE Non-Dispersive Infrared Analyzer) sensitive to both carbon dioxide and methane. The
analyzer can be periodically calibrated using the calibrated gas mixture to ensure accuracy.
The calibrated gas tank contained a mixture of gases in one of the ratios listed in Table 2.3
depending on the specific experiment being performed. The gas mixtures were provided by
12
Table 2.1: Manufacturer listed properties of evaluated carbon dioxide sensors
Sensor Supplier Type Sampling Method Cal. Range Op. Range
K-30 SE-0018 CO
2
Meter NDIR flow or diffusion 0-5000 ppm 0-10000 ppm
COZIR AMB GC-020 CO
2
Meter NDIR flow or diffusion 0-5000 ppm 0-10000 ppm
Gascard CO
2
GHG Analytical NDIR flow 0-50000 ppm 0-50000 ppm
MSH-P/CO2/NC/5/V/P/F Dynament NDIR diffusion 0-2491 ppm 0-5000 ppm
MSH-DP/HC/CO2/NC/P/F Dynament NDIR diffusion 100-2500 ppm 0-5000 ppm
Telaire T6615 General Electric NDIR flow or diffusion 0-2000 ppm 0-2000 ppm
Sensor Warm Up T Humidity Auto-cal V Input Avg. I Peak I
K-30 SE-0018 <1 min 0-50
◦
C 0-95% Yes 4.5-14 VDC 40 mA <150 mA
COZIR AMB GC-020 <3 s 0-50
◦
C 0-95% Yes 3.25-5.5 VDC 1.5 mA 33 mA
Gascard CO
2
30 s 0-45
◦
C 0-95% Yes 7-30 VDC
MSH-P/CO2/NC/5/V/P/F 45 s -20-50
◦
C 0-95% No 3.0-5.0 VDC 75-85 mA
MSH-DP/HC/CO2/NC/P/F 45 s -20-50
◦
C 0-95% No 3.0-5.0 VDC 75-85 mA
Telaire T6615 10 min 0-50
◦
C 0-95% Yes 0-5 VDC 33 mA 180 mA
Table 2.2: Manufacturer listed properties of evaluated methane sensors
Sensor Supplier Type Sampling Method Cal. Range Op. Range
MQ-4 Futurelec Chemoresistive diffusion 200-10000 ppm
Gascard CH
4
GHG Analytical NDIR flow 0-50000 ppm 0-50000 ppm
MSH-P/HC/NC/5/V/P/F Dynament NDIR diffusion 0-5000 ppm 0-10000 ppm
MSH-DP/HC/CO2/NC/P/F Dynament NDIR diffusion 5000-11000 ppm 0-10000 ppm
TGS-2600 Figaro Engineering Chemoresistive diffusion 1-30 ppm
TGS-2610 Figaro Engineering Chemoresistive diffusion 1000-25000 ppm
TGS-2611 Figaro Engineering Chemoresistive diffusion 500-10000 ppm
Sensor Warm Up T Humidity Auto-cal V Input Avg. I Peak I
MQ-4 No
Gascard CH
4
30 s 0-45
◦
C 0-95% Yes 7-30 VDC
MSH-P/HC/NC/5/V/P/F 30 s -20-50
◦
C 0-95% No 3.0-5.0 VDC 75-85 mA
MSH-DP/HC/CO2/NC/P/F 30 s -20-50
◦
C 0-95% No 3.0-5.0 VDC 75-85 mA
TGS-2600 No 5.0±0.2 VDC 4.2±4 mA
TGS-2610 No 5.0±0.2 VDC 5.6±5 mA
TGS-2611 No 5.0±0.2 VDC 5.6±5 mA
Sensors with no listed warm up time, but required 7 day burn in time
and certified by Airgas Inc. The calibrated gas mixture diluted using either air or nitrogen
gas by a set of flow controls to produce specific partial pressures of analyte gases. In the
case of sensors configured for gas flow, the sensor being tested was connected directly to
the apparatus by flexible hose connection. For sensors that are diffusion based, the sensor
was placed in an exposure chamber connected by flexible hose. Using the exposure chamber,
the relative rise time of the various sensors was measured and compared. The mixing time
or the time required for the chamber to reach and maintain gas concentration, even when
the concentration of the incoming gas was changing was determined to be negligible with
13
respect to the response times of the sensors. All experiments were referenced to the California
Analytical Instruments ZRE Analyzer.
Table 2.3: Ratios of calibrated gases used in mixed gas experiments.
Carbon Dioxide Methane Bulk Gas
3000 ppm 3000 ppm nitrogen
100 ppm 100 ppm nitrogen
0 ppm 20 ppm nitrogen
Commercially available carbon dioxide and methane sensors were selected and evalu-
ated based on the needs for a potable self-powered sensor. Table 2.1 lists the selected carbon
dioxide sensors with important properties obtained from the manufacturer. Table 2.2 lists the
methane sensors and respective properties. Certain sensors were chosen with specific provided
specifications. The K-30, COZIR, Dynament, and Telaire sensors are all NDIR sensors. These
were chosen as low-cost, lightweight sensors with satisfactory detection parameters of carbon
dioxide. Dynament also provides a dual gas NDIR sensor designed to measure both carbon
dioxide and methane concentrations. The carbon dioxide and methane Gascard sensors sold
by GHG Analytical were an order of magnitude more expensive than the other NDIR sensors,
having a cost between that of the low-cost sensors and that of the bench-top analyzers.
Their specifications and the included internal pressure and temperatures compensation make
them attractive enough to make up for the expense. In addition to the Gascard sensor,
the Dynament hydrocarbon sensors were chosen as the inexpensive candidates for methane
detection. Chemoresistive sensors include the MQ-4 and TGS-26xx series sensors. The TGS
sensors manufactured by Figaro Engineering Inc. are used in commercial methane detectors
and have been previously evaluated by other groups [
82
]. There are several different MQ
versions optimized for hydrocarbon sensing. The MQ-4 from Hanwei Electronics was chosen
as this variant was specifically tuned for methane. The chemiresistive sensors required various
minimum conditioning periods before use by their respective manufacturers. This “burn-in”
time was met or exceeded for all chemiresistive sensors.
Due to the unique interface requirements of each sensor unit, development kits
were purchased when possible. For those without development kits available, an Arduino
microcontroller with a prototyping board was used. The chemiresistive sensors were energized
using the recommend voltage for the heater element and the response was measured as
14
Carrier
Gas
(N or Air)
2
Calibrated
Gas
Mixture
CAI Inc. ZRE
VENT
VENT
Needle Valve
T - Connector
Solenoid Valve Flow Controller
Mineral Oil Bubbler
(blowoff vent)
Figure 2.1: Component diagram of controlled gas exposure apparatus with chamber for diffusion-type
sensors. The gas from the cylinders is mixed by controlling the ratio of flow by the flow controllers. Multiple
solenoid valves are able to completely shut flow off from the tanks. The mixed gas is fed to the flow chamber.
The gas flowing through the chamber exits and goes to the California Analytical Instruments Inc. ZRE unit
for analysis. A mineral oil bubbler is present to prevent overpressure of the exiting gas.
voltage, after buffering and filtering, across a 10 kΩ resistor. Data were logged as a function of
time using software provided by the manufacturer as part of development kits when provided,
or using an Arduino microcontroller with microSD card.
15
2.3.1 Precision and Baseline Noise Tests
The measurement precision of each sensor was performed using the exposure appa-
ratus discussed above. The baseline carbon dioxide tests were performed with a measured
concentration of approximately 400 ppm in nitrogen gas, while the baseline methane tests
were performed under medical grade air. The 0 ppm methane concentration was chosen as the
expected atmospheric or environmental level will be close to 0 ppm. All of the chemiresistive
sensors required atmospheric concentrations of oxygen to measure the methane. The concen-
tration of gas used to fill the chamber was verified with the bench-top analyzer (California
Analytical Instruments, Inc. ZRE Non-Dispersive Infrared Analyzer). The precision of the
sensors was determined by a 20 to 30 hour data collection time at a known concentration
and uniform flow. First, a Fourier analysis was performed on the data collected from each
sensor. These analyses showed no significant periodic variations in the output during the
respective tests. Next, the data streams were processed to provide the individual difference
from the mean reported value, and a histogram of these differences was plotted. The digitized
sensor outputs have a finite number of possible output values, so no further bins were created
while producing the analysis. A Gaussian fit was applied to the histograms to determine the
degree of fit. The histograms with Gaussian peak fitting can be seen in Figures 2.4 and 2.5
for the carbon dioxide and methane sensors, respectively. These plots have been scaled to
show detail. The narrow FWHM of the K-30 and Gascard sensors plot in Figure 2.4 shows
that both devices can be shown to give high quality data near the global mean concentration
of carbon dioxide. Similarly, the Gascard in Figure 2.5 shows good data quality near the
global baseline for methane. The digitization level of the Dynament sensors causes errors
near this methane concentration. No curve could be reasonably obtained from one of these
sensors due to it only oscillating between two values. The MQ-4 sensor is notably out of scale
from the other methane sensors. The adjusted scale in Figure 2.5-B allows this peak to be
visible. This clearly shows that the MQ-4 should not be used for quantitative work. Finally,
the concentration drift was measured using the difference between the highest and lowest
points in an extended run.
16
580
600
620
640
660
680
700
0 100 200 300 400 500 600
Gascard
time(hr)
400
420
440
460
480
500
520
K30
320
340
360
380
400
420
440
DynamentDual
Figure 2.2: Long term baseline data were collected at the atmospheric baseline conditions (using a bottle of
compressed medical grade breathing air, approximately 400 ppm carbon dioxide) in the gas mixing chamber.
The baseline drift and noise can be observed in these plots.
2.3.2 Limit of Detection
Utilizing the gas mixing apparatus described in Section 2.3, a series of tests were
conducted to determine the limit of detection and calibration curve of the selected sensors.
For these tests, multiple sensors were in operation concurrently as shown in Figure 2.6. The
sensors are organized by flow capable sensors such as the Gascard units and environmental
exposure-based sensors housed within the gas enclosures. Sensor modules in the small gas
flow enclosures included the K30 carbon dioxide sensor and socket for chemiresistive methane
sensors (MQ-4 and the TGS series). The larger chamber contained the COZIR, Telaire,
Dynament, and K30 sensors.
A calibration curve of each sensor was developed using varied concentrations of gas
17
0
5e-07
1e-06
1.5e-06
2e-06
0 10 20 30 40 50 60
TGS2611 conductivity (S)
time (hr)
1.2e-05
1.25e-05
1.3e-05
1.35e-05
MQ4 conductivity (S)
Figure 2.3: Long term baseline data were collected at the atmospheric baseline conditions (using a bottle
of compressed medical grade breathing air, approximately 0 ppm methane) in the gas mixing chamber. The
baseline drift and noise can be observed in these plots.
through system and, when appropriate, subtracting the baseline reading to the average
concentration after stabilization. In these experiments, the carrier gas was first introduced.
After a stable baseline was obtained, the calibrated concentration was introduced, after 24
hours, the calibrated gas was introduced, and after 24 hours, it was turned off. The carrier gas
was reintroduced and the system was allowed to stabilize for the next measurement. During
this procedure, contraction data from each sensor was continually collected. With this method,
the average baseline and response to each concentration can be extracted. This process allows
the ramp-up when the calibration gas is introduced and ramp-down when the calibration gas
was vented, along with any initial overshoot as observed for the chemostresistive sensor to be
edited from the data stream. The California Analytical Instruments Inc. ZRE Non-Dispersive
Infrared Analyzer was used design calibration curves for each sensor. The mean (
µ
) and
standard deviation (
σ
) were calculated from each individual baseline and average at each
calibration point. The Limit of Detection (
L
D
) was calculated for each sensor run using the
standard approach [83, 84]:
18
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
-40 -30 -20 -10 0 10 20 30 40
Probablity
DifferencefromMean(ppm)
K30(scalefactor4.35)
COZIR(scalefactor2.17)
Gascard(scalefactor15.47)
DynamentCO(scalefactor16.43)
DynamentDual(scalefactor4.53)
Telaire(scalefactor10.09)
Figure 2.4: This is a plot of the probability density function by Gaussian non-linear fitting on frequency
distribution of responses by the carbon dioxide sensors. Responses for each sensor were counted in bins based
on distance from the mean in concentration units. The Gaussian peaks have been scaled in the y range to
show detail. Plots have been scaled such that the most probable point is set at 90% probability. The scale
factor for each curve is listed on the legend.
L
D
= 3 · σ
Plots were made of the sensor response at different carbon dioxide spikes. These
figures were produced such that:
x = µ
ZRE,peak
− µ
ZRE,baseline
y = µ
sensor,peak
− µ
sensor,baseline
.
An example plot is shown below in Figure 2.7.
19
The Beer-Lambert law results in all sensors with optical detection schemes showing a
linear response. The chemiresistive sensors produce non-linear calibration curves. As these
sensors are dependent on the analyte gases adsorbing to the surface of a substrate to generate
an electrical signal, the response curve is can be modeled using a Langmuir-like or Langmuirian
form. This form is consistent with that provided by the manufacture. The calibration curve
was determined by fitting the experimental data using the Levenberg-Marquardt algorithm
in gnuplot [85] taking the form:
f(x) =
Q · b · x
1 + b · x
A trendline is shown in both parts of Figure 2.7. The slope (
m
) and intercept (
b
) are
used in calculations from the results of linear regression modeling for optical sensors, and the
coefficients of the Langmuir isotherm
Q
and
b
are used in calculations from the results of
fitting the isotherm model. The mean limit of detection (
µ
L
D
)was adjusted for each sensor.
Since detection limits are not comparable across arbitrary and unrelated units, these values
were correlated to the ZRE Non-Dispersive Infrared Analyzer response by the calibration
curve produced in the trendline. For optical sensors with a linear calibration curve:
L
D(corr)
= m · L
D
+ b
For sensors fit to a Langmurian curve:
L
D(corr)
=
Q · b · L
D
1 + b · L
D
2.4 Results and Discussion
The tests showed that the Edinburgh Gascard sensors reliably generated accurate
information on the gas concentration. The GE Telaire and K-30 sensor both gave reasonable
results for sensors in their price range. The Dynament sensors and CO
2
Meter COZIR sensor
were found to be less reliable than other sensors in their cohort. Finally, the MQ-4 sensor
proved to be rather inaccurate for gauging gas concentration, rather it is more suitable to
giving a rough idea of the presence of methane.
20
2.4.1 Precision and Baseline Noise Tests
The summary of results from this analysis can be found in Tables 2.4 and 2.5. Statistical
analysis data for these curves are given in terms of standard deviation and the full width
at half max (FWHM). Of the carbon dioxide sensors show in Table 2.4, the Gascard and
K-30 sensors both showed relatively narrow distributions of response. The Gascard shows a
strong degree of fit with the Gaussian plot, giving a FWHM of 5.01 ppm. The K-30 tends
toward higher concentration values and appears to show two overlapping peaks. These peaks
have similar FWHM values, suggesting a shift in the mean measured value. The ppm values,
being the ratio volume of the anatyte to a unit volume, are dependent on temperature and
pressure. Possible periodic temperature and pressure variations, which are corrected for by
the Gascard sensor, are responsible. It is suggested that implementation using the K-30 sensor
for high-accuracy measurements should incorporate pressure and temperature sensing units
to correct these deviations in post-processing. The breadth of the strongest Gaussian peak
in the histogram of responses from the K-30 sensor suggests a similarly narrow distribution
around the mean as the Gascard. For both of these sensors, the drift was
±
20 ppm. The
distribution of responses around the mean by the GE Telaire sensor produces a diffuse broad
distrubution compared to the relatively narrow on from the K-30 and Gascard sensors. The
breadth of the GE Telaire peak (FWHM 10.4) is interesting when compared to the K-30
sensor, as the sensing mechanism and path lengths are very similar.Looking closer, it was
discovered that the Telaire sensor is sensitive to ambient light levels. The data reported this
sensor shifting noticeably with ambient light levels. The change in response from the K-30
sensor under similar conditions is negligible.
Comparing the response for both the Telaire and K-30 sensors was monitored with and
without ambient light. The data showed that the Telaire sensor is significantly more sensitive
to outside light sources than the K-30. The change in response from the K-30 sensor during
this test was negligible, while the data reported by the Telaire was observed to shift noticeably.
The COZIR AMB GC-020 sensor compared with the K-30 sensor (both produced by the
same manufacturer) was significantly nosier and less precise. Of all carbon dioxide sensors
tested, the sensors produced by Dynament showed poor performance at typical atmospheric
concentrations. While the single-gas carbon dioxide sensor gave a level of precision on par
with the COZIR sensor, the dual-gas sensor reported carbon dioxide concentrations over a
21
wide span of values. The Dynament sensors can perform acceptably when utilized for hazard
monitoring, but they are less suitable for gas monitoring.
Of the methane sensors shown in Table 2.5 the Edinburgh Gascard again proved the
most precise. The normal distribution of responses around the reported mean from the
Gascard is a very narrow peak. The Gascard did report some relatively large fluctuations,
several peak widths higher than the mean response. Although these fluctuations could be
explained by the exposed nature of the sensor during these tests, it suggest that some digital
filtering may be required for portable devices. The single gas Dynament methane sensor
produced a narrow distribution of responses, but the results from the dual-gas Dynament
sensor were inconclusive. The dual-gas Dynament sensor only reported two values for methane,
rather than a normal distribution of several values. Possibly, the internal digitization in
the sensor provides limited resolution around the baseline value of 0 ppm. Assuming that
an approximate FWHM could be approximated by the difference between these two points
around a mean, the dual-gas sensor would have a FWHM equivalent to that produced by the
single-gas sensor. Finally, the MQ-4 sensor produced a wide range of values. The distribution
around the mean is so broad, that the tails are off the scale in Figure 2.5. While the FWHM
of these responses from the MQ-4 show that the peak only deviates 1.19 ppm around the
mean, it is comparatively much less precise than the other sensors in the experiment.
With the exception of the carbon dioxide and methane Gascard sensors, there was
poor accuracy. In the case of some sensors including the K-30, Telaire, COZIR, and Dynament
carbon dioxide sensors, this issue could be rectified by recalibration of the sensors. All of
these sensors must be calibrated before use and cannot be used directly from the supplier if
the user hopes to obtain an accurate concentration value. For some of the tested devices such
as the MQ-4 methane sensor, good accuracy may be outside of the reasonable expectation of
sensor ability.
2.4.2 Response Time
Most of the sensors tested registered change in concentration on a timescale which
could not be measured accurately with the current experimental design. For the optical
sensors, the sampling time was controlled by internal electronics, which sample several times
a minute. For environmental monitoring, the response from all sensors were effectively
22
Table 2.4: Precision and Accuracy of Carbon Dioxide
Sensors
Sensor
¯
C
*
σ FWHM
K-30 SE-0018 451 1.91 4.51
COZIR AMB GC-020 362 14.1 33.3
Gascard CO
2
635 2.12 5.01
MSH-P/HC/CO2/ 395 86.4 204
MSH-DP/CO2/ 2316 17.6 41.4
Telaire T6615 454 4.42 10.4
Table 2.5: Precision and Accuracy of Methane Sensors
Sensor
¯
C σ FWHM
MQ-4
†
97.6 0.507 1.19
Gascard CH
4
15.9 0.580 1.37
MSH-P/HC/CO2/ 48.2 3.54 8.33
MSH-DP/HC/ 15.4 no data no data
*
While
¯
C
denotes the average concentration as de-
termined by the sensor during the test, it should
be noted that these data were collected by un-
calibrated sensors. The values of
σ
and FWHM
are based on
¯
C.
†
Data for the MQ-4 are listed in terms of electrical
response rather than parts-per notation.
‡
All other data in Tables 2.4 and 2.5 are in units
of ppm
immediate. However, chemoresistive sensors responded on a notably different timescale
was the MQ-4 sensor. In these experiments, the MQ-4 methane sensor was selected as a
worst-case model sensor. A smaller sensor enclosure (internal volume 1 in
3
) was constructed
to maximize gas exchange within the chamber and minimize the time required to obtain a
stable new concentration. A test was conducted to determine the response time of this sensor
by introduction of gas to the system until a stable value was produced, then vented with
purified air until the previous baseline was regained. The ratio of gas introduced to the system
was changed between each acquisition period to determine if there was a relationship between
the concentration of introduced gas and the time to produce a stable response. Figure 2.8
depicts a plot of the responses from one of these experiments. The sensor produces a sharp
overshoot, with some instances displaying a percentage overshoot exceeding 100% of the final
23
concentration determined by the system. This large overshoot is likely due to the heating coil
inside the sensor, which artificially inflates the apparent concentration as voltage is flowed
through the resistive wire. Minor signal ringing follows this spike, until the response is made
stable. A smaller overshoot can be observed when the mix gas is shut off, and replaced by
pure air. The MQ-4 works by a thermometric response, and as such, it is reasonable that
the mixed gas introduced to the system changed the temperature near the sensor, and this
contributed to the degree of overshoot observed. This would not explain the overshoot when
the gas was switched to pure air, as it was the same temperature as the mixed gas.
The rise time of a system, defined as the time for the response to change from 10%
of the mean response to 90% of the mean response, is less useful for this system due to the
extreme overshoot observed. Settling time, the time for a system to change from the baseline
to a window around the mean during the ringing portion, is more useful for this application.
The settling time for the MQ-4 sensor did not appear to be concentration dependent. A
stable value was produced within 2.5% of the mean after 78
±
10 s, averaged over all response
time experiments.
2.4.3 Limit of Detection
The correlated limits of detection are listed in Table 2.6. Of the tested carbon
dioxide sensors, the Edinburgh Gascard shows the greatest sensitivity. This is consistent
with expectations based on the path length of the sensor. The Dynament single analyte
sensor shows a lower limit of detection than the dual analyte sensor produced by the same
manufacturer. The K-30 and Telaire sensors show comparable detection limits. The COZIR
sensor exhibits the poorest limit of detection.
Of the tested methane sensors, the Edinburgh Gascard again shows the greatest
sensitivity. The MQ-4 and TGS2611 sensors, have very low limits of detection. It was
observed that these thin film sensors were very sensitive to changes in temperature and
humidity. It should be noted that these tests were carried out in a controlled environment
which sought to eliminate those sources of error. An application without these sorts of
controls may be subject to much more error. Furthermore, individual MQ-4 sensors varied a
great deal in response compared to other sensors of the same type. The Dynament sensors
have appear to have a limit of detection under 10 ppm. These numbers are artificially deflated.
24
Table 2.6: Detection Limits of Carbon
Dioxide Sensors
L
D(corr)
K-30 SE-0018 31.1
COZIR AMB GC-020 65.7
Gascard CO
2
0.00862
MSH-P/HC/CO2/ 57.6
MSH-DP/CO2/ 6.60
Telaire T6615 22.9
Table 2.7: Detection Limits of
Methane Sensors
L
D(corr)
MQ-4 1.24
†
Gascard CH
4
0.569
MSH-P/HC/CO2/ 7.44
‡
MSH-DP/HC/ 2.42
‡
TGS-2600 33.3
†
TGS-2610 29.0
†
TGS-2611 3.95
†
†
Calculated from Lang-
muirian fitting.
‡
Failed to produce consistent
response within tested range.
The Dynament sensors could not reliably detect any concentration of gas below 100 ppm, and
these null results have skewed the trendline of detection limits. Values we observed cannot
be guaranteed below this threshold. The TGS-2600 and TGS-2610 sensors are generalist
hydrocarbon and contaminant sensors, these sensors were less sensitive to changes in methane
than the TGS-2611, which is specific to methane.
2.5 Conclusion
Based on the comparison of these commercially available low-cost sensors, selection of
a sensor for practical applications can be simplified. The obvious choice for low concentration
applications is the Gascard series of sensors. However, this sensor is comparatively more
25
expensive than the other sensors in this paper on a scale of 10-100 times. For applications
with modest budgets, the K-30 and Telaire sensors are viable for carbon dioxide detection,
and the TGS-2611 sensor is similarly viable for methane. The K-30 sensor costs roughly a
tenth of the cost of the similarly-able Telaire sensor. Therefore, we also recommend it over
the Telaire offering. The MQ-4 is an incredibly inexpensive alternative to the TGS series,
available for only a few dollars. For very low cost devices, the MQ-4 sensor is an option if the
user only needs a rough estimate of concentration, rather than the precision found in other,
more expensive, offerings. During our study, the TGS sensors proved difficult to purchase,
most suppliers not stocking this device and only at a high price when available. Use of the
Dynament sensors is not advisable due to inaccuracies and inconsistencies of the devices
considering their cost (roughly 100 times that of the MQ-4 sensor).
Sensor choice for the detection of ppm level concentration of carbon dioxide and
methane is a matter of compromises. There appears to be no single sensor which simultaneous
fulfills traits of sensitivity at atmospheric levels, low power consumption, low price, and
market availability. It is the opinion of the authors that the devices which best fill these
needs from the available options are the K-30 and MQ-4 sensor. The K-30 sensor provided
acceptable sensitivity for a low cost in terms of money and power. The MQ-4 sensor appears
to be surprisingly sensitive to small changes in concentration (based on the adjusted limit
of detection discussed in Table 2.7), but data it produces should not taken as accurate (see
the discussion of the broad peak in Figure 2.5-B). However, the MQ-4 does appear to be
able to detect larger changes in concentration which would represent localized leak. We must
consider the application the sensor array. If the cost an accurate sensor such as the methane
Gascard is 1000 times greater than an inaccurate sensor such as the MQ-4, it is more valuable
to the goals of the network to have a much larger number of units than it is to have just a
few powerful sensors. If the goal of the device array is to detect a leak, the MQ-4 is sufficient
for most units. We believe that the TGS2611 would be more suited to this task, but the
difficulty in acquisition prevents it from being a viable option at scale at this time.
26
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
-0.02 -0.01 0 0.01 0.02
Frequency
DifferencefromMean
MQ-4(scaledfactor1.44)
Gascard(scalefactor13.06)
Dynament(scalefactor1.08)
DynamentDual(nofit)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
-10 -5 0 5 10
Frequency
DifferencefromMean
MQ-4(scaledfactor1.44)
Gascard(scalefactor13.06)
Dynament(scalefactor1.08)
DynamentDual(nofit)
Figure 2.5: Figure 2.5-A is a plot of the probability density function by Gaussian non-linear fitting on
frequency distribution of responses by the methane sensors. Responses for each sensor were counted in bins
based on distance from the mean in concentration units. The Gaussian peaks have been scaled in the y
range to show detail. Plots have been scaled such that the most probable point is set at 90% probability.
The scale factor for each curve is listed on the legend. The digitized output from the Dynament Dual sensor
produced only two possible values. A Gaussian fit could not be calculated from the available points. The
MQ-4 extent curve is not visible in the domain of Figure 2.5-A. By increasing the scale of the domain to
that in Figure 2.5-B, the curve is visible.
27
CH
Gascard
CO
Gascard
2
4
Control
Board
Control
Board
Control
Board
Control
Board
Gas Mixer
Gas Analyzer
Sensor Flow Box
Sensor Flow Box
Figure 2.6: The gas mixing apparatus discussed in Section 2.3 was used with additional sensor housings.
The diagram shows the flow of gas from the mixing apparatus to Gascard sensors, small enclosures containing
sensors, the large gas flow chamber, and finally to the ZRE gas analyzer.
28
0
500
1000
1500
2000
0 100 200 300 400 500 600 700 800 900 1000
SensorResponse
ZRE(ppm)
non-linearfit
thermaldetectordata
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
SensorResponse
linearfit
opticaldetectordata
Figure 2.7: A sample plot of the calibration of sensor signals to ZRE Non-Dispersive Infrared Analyzer
reported concentration during carbon dioxide and methane spikes. The top part of the figure shows the
response of an optical carbon dioxide sensor with the expected linear behavior. The bottom part of the
figure shows the response of a chemiresistive methane sensor with a fit of the non-linear response.
29
0 5 0 0 1 0 0 0 1 5 0 0 2 0 0 0 2 5 0 0
0
2 0 0
4 0 0
6 0 0
8 0 0
1 0 0 0
S e n s o r R e s p o n s e
T i m e ( s )
Figure 2.8: The MQ-4 sensor showed noticeable delay between introduction of gas and production of a
stable response. In this plot, the large overshoot during sensor cycling, likely caused by the heating coil in
the device, is apparent.
30
Chapter 3
Determining the Battery Shutoff Circuit Behavior of a Commercially Available Solar
Charging Circuit
3.1 Introduction
The nature of the sensor array designed to cover a relatively large area of land imposes
limitations upon the power supply of the system. Since the area to be covered may include
elevation changes, anthropogenic developments, and other uncontrollable factors, it would
not be convenient to constrain the array by wired connection as the array will be adapted
to individual site conditions. Therefore, each unit must be designed with internal and
self-sustaining power supplies.
Two different types of commercially available solar power generation and storage units
with weatherproof enclosures were selected from Tycon Power Systems. The majority of the
sensors would be housed in the Remote Pro
2.5 W
Continuous Remote Power System die
cast enclosure and a few sensor units with features requiring more power were housed in the
Remote Pro
15 W
Continuous Remote Power System steel enclosure . The
2.5 W
enclosure
includes a
12 V
battery rated for
9 A h
of use, a charging and distribution circuit, and a
10 W
solar panel. The
15 W
enclosure includes a
12 V
battery rated for
98 A h
, a charging and
distribution circuit, and a 60 W solar panel [86].
Before equipment construction, the power shutoff features of the Tycon systems
required evaluation. Tests were undertaken to determine battery behavior drain while
connected to the proprietary circuit provided with the Tycon units.
31
3.2 Discharge Behavior and Cutoff Point
The
9 A h
battery specification was tested for veracity by monitoring a discharge
of power through a resistor connected to the supply terminals on the distribution circuit
included with the enclosure. The voltage through the circuit was monitored by a PicoScope
4224 oscilloscope to determine the time and value when the circuit was opened to prevent
battery damage. The tests put minimal strain on the battery, in regard to the specified
maximum number of charging cycles. Multiple batteries were tested, allowing charging while
a test took place, and the shut down time was averaged across batteries.
The circuit ceased to discharge after the voltage through the circuit dipped below a
10.7 V
threshold value. This number is close to the expected threshold value, which is labeled
on the circuit. The behavior of the circuit near low voltage levels in one test is shown in
Figure 3.1 in terms of lifetime with respect to the voltage detected. While the battery begins
naturally losing stored energy at an increased rate near the end, the protection circuit cuts
in at about 10.7 V and breaks the circuit.
3.3 A Direct Comparison of Batteries and Damage Due to Repeated Discharge
The value of the resistor in the simple circuit varied over the course of the experiment.
Individual shutoff behavior at certain power draw rates is shown in Figures 3.2, 3.3, and 3.4.
While these figures are summarized in Figure 3.5, the individual endpoints show the types
of deviations. Notably, the shape of the curve near the cutoff point is different for the two
batteries. One ends in a simple pseudo-parabolic curve, while the second battery seems to
resist the threshold more, taking on a pseudo-Gaussian shape. This interesting effect does
not seem to have apparent correlation to the capacity of either battery, it is interesting to see
two different curve shapes for batteries which should be similar in behavior.
Tests showed that for these batteries and this circuit, the time which battery took
to discharge was inversely related to the current allowed to flow through the circuit (see
Figure 3.5). This is behavior which is expected for a normal battery capacity. Since more
than one battery was used in this test, the capacity of the two batteries can be compared.
Furthermore, the battery capacity of the battery changed as the experiment was
repeated multiple times. In Figure 3.6 the calculated capacity of each battery after each test
32
Figure 3.1: Battery shutoff behavior by protection circuit.
is shown. These data suggest that the battery must be decaying with each cycle, despite
the intervention of the shutoff circuitry. Based on these data, the capacity of the analyzed
12 V
battery at the beginning of the tests was determined to be
5.5 A h
and
5.1 A h
for the
first and second batteries, respectively. This is well below the specified capacity of
9 A h
by
the distributor. One possible explanation for this behavior is that the power distribution
circuit provided by the enclosure supplier does not adequately protect against memory effects.
It was noted during production that batteries plugged into the power distribution circuit,
without being connected to recharging circuits or loads, would discharge over time to an
unusable state. This required many batteries to be replaced, as the full discharge left them
below the power threshold required to charge on a simple trickle charger.
33
Figure 3.2: Shutoff behavior of battery circuit with a 141 Ω load.
3.4 Conclusions
Based on the tests conducted on the batteries connected to the shutoff circuit during
prototyping, we can determine a few things. First, while the nominal capacity of the batteries
is listed by the manufacturer as
9 A h
, the true capacity is closer to
5.35 A h
. This reduces
the available uptime for future systems based on this equipment. Second, the protection
circuit included with the Tycon equipment does not prevent cycling damage to the battery.
Although the protection circuit can operate as designed to prevent the battery from dropping
below a threshold where significant damage to the capacity may occur, it should not be
recommended to get to that point. Systems connected to this device should be designed to
enter a low-power mode before this threshold. Finally, batteries should be kept disconnected
while not in service. Although lead-acid batteries are designed to be rechargeable, the trickle
charge circuit of the Tycon system cannot be relied on to recharge a battery which has been
allowed to completely discharge.
34
Figure 3.3: Shutoff behavior of battery circuit with a 188 Ω load.
Figure 3.4: Shutoff behavior of battery circuit with a 235 Ω load.
35
0
50000
100000
150000
200000
250000
300000
350000
400000
0.05 0.055 0.06 0.065 0.07 0.075 0.08 0.085 0.09
Time to Shutdown (s)
Current (A)
Average of Batteries Tested
Maximum Batteries Tested
Minimum Batteries Tested
Figure 3.5: The time for a battery to reach the discharge cutoff point was determined by varying the
resistor loads. We can see from this plot that this is a linear relationship.
36
0
1
2
3
4
5
6
7
8
1 2 3 4 5 6 7 8
Capacity (Ah)
Test Number
Battery 1
Battery 2
Figure 3.6: Changes in the battery capacity over time suggest that the batteries quickly decay over a few
cycles. The capacity of two batteries are plotted here with respect to repeated use. The tested batteries
were shut down at 10.7 V like all other battery tests. This plot shows that the capacity of the batteries still
depleted with repeated discharges to this point.
37
Chapter 4
Networking with from an Xbee Modem Mesh to a Cellular Modem
4.1 Network Topologies for Remote Sensing
The design of a network is a much-studied endeavor, and many of the nuances of
network topologies have been described in literature [
87
,
88
]. It is useful to examine each of
these topologies to determine which is best suited for the task at hand. In the case of device
designed for remote monitoring must be able to provide the data to researchers at a distant
location. Additionally, as this is a network of multiple sensing units, there must be some
way to communicate data between sensors. There are several methods to do this. The most
straightforward method of designing a network, is to establish point-to-point communications.
To field this method, each node in the network would utilize a long range modem to connect
to the ‘home’ server, as depicted in Figure 4.1. Outfitting each node with a powerful modem
would be costly in parts, network plan maintenance, and energy. The infrastructure of the
units would require large power sources to keep up a high duty cycle. Thus, it is practical
to limit network connections directly to the server by the large modems to as few units as
possible. In this manner, two classes of nodes are established: communication nodes and
sensor nodes. The sensor nodes could use a lower power communication method.
The most cost effective option for sensor connection is a wire to transmit signals
between units, yet this is inadequate for this project. Wired connection is not used due
to the scope and shape of the network. Simple wired connections often follow some basic
network topologies. The most rudimentary wired connection would connect each unit to
38
Figure 4.1: This cartoon depicts the nodal arrangement of a point-to-point network. Each unit connects
directly with the ‘home’ server.
the nearest neighbor. This line network topology, as depicted in Figure 4.2, requires each
unit to pass information to the nearest neighbors before reaching a terminal node, which
would be equipped with a large modem to send data to the ‘home’ server. A variation of
this design, is the bus network topology, shown in Figure 4.3. The bus allows for a slightly
more broad area to be managed by a network, with addressing functionality to send data to
correct locations. Both of these networks do not match the needs of this project. A wired
connection produces a physical failure point which is difficult to maintain for prolonged
periods in outdoor conditions. The scope and breadth of the project demands that a large
area be covered by sensor, rather than a long line. To wire a line or bus network over an
area, would require the snaking of data cabling around and possibly over difficult terrain. A
wireless network is much better suited to this task.
Figure 4.2: This cartoon depicts the nodal arrangement of a line or linear network. Data are passed
between nearest neighbor units in a line until they reach the ‘home’ server.
A wireless network can offer a large number of network topologies without a tangle of
wires. A simple network design which can cover a large area as intended by the project is a
star topology, depicted in Figure 4.4. This network directs all data to a central unit. We
can design this central unit to have a longer range modem to connect to the ‘home’ server.
Yet this design is also limiting. By requiring all sensor nodes to communicate directly to
a communication node, the network reaches a size limit based on the signal range of the
antenna. For low power wireless modems, this range limit may be too small. To address
39
Figure 4.3: This cartoon depicts the nodal arrangement of a bus network. Data are passed in a bus cable
to the ‘home’ server. Units are assigned addresses. If data packets match the address the data are accepted
and stored.
this shortcoming, it becomes beneficial for all units in the network to communicate data
to the others. The theoretical example of this is the mesh network, depicted in Figure 4.5.
In this archetypal example, each node in the network connects with every other node. In
this way, a single unit can be designated as a communication node. When the sensor nodes
are designed to pass data between them until the data reach the communication node. The
constraints of the mesh network are great, requiring each unit to be within range of the other
units. Instead, a hybrid network is desirable, see Figure 4.6. This network design allows as
many of the nodes to mesh as possible, while keeping the ability to adapt forms of the star,
bus, and linear topologies. This network design provides the most adaptability for an area
network to conform to the needs of the layout of units and the environmental conditions.
4.2 Modems and Radios
To construct a hybrid network for our sensor array, we need two types of wireless
radios. The mesh portion of the network is performed using the Xbee mesh protocol. The Digi
International XBee-PRO 900HP module XBP9B-DMUT-002 was used for this application
(Figure 4.7). This unit is a spread spectrum transmitter licensed to operate between 902.4
MHz and 927.6 MHz at an output of 0.298 W [
89
]. The device communicates with a controller
via UART interface and can transmit to other units via the DigiMesh, repeater, point-to-pint,
point-to-multipoint, and peer-to-peer protocols. The unit selected transmits data at a rate of
40
Figure 4.4: This cartoon depicts the nodal arrangement of a star network. Data are passed from low level
units to a higher level unit at the center.
200 Kbps with a claimed maximum range of 9 miles with the proper antenna. For this setup,
we will be using small omni-directional one-eighth wavelength dipole antenna connected
by U.FL cable to the modem. The modems in the network will communicate using the
Digimesh protocol. Digimesh is a a proprietary implementation of the Zigbee protocol which
homogenizes the network, and Zigbee is an enhancement of the IEEE 802.15.4 protocol [
90
]
which adds mesh topology to the existing physical address and mac address protocols. The
Zigbee protocol adds routing, ad hoc networking, and self-healing mesh capabilities to the
established protocol [
91
,
92
]. This addition allows the units to automatically self-assemble a
mesh of units, designating each as a coordinator, router, and end device.
Each part of the Zigbee mesh has a particular job, the coordinator is a central unit
that assigns addresses, the routers move data to the coordinator, and the end devices just
supply data. This topology allows the end devices to persist in sleep mode for most of the
time, waking up only to sample data and send it along. However, it requires the router and
coordinator nodes to remain more active. This shortcoming is amended with the use of the
Digimesh topology. This protocol allows for the formation of a true mesh which does not
41
Figure 4.5: This cartoon depicts the nodal arrangement of a mesh network. Data are shared between
units equally, and all units are connected equally. Any node can be replaced with a communication node to
transmit data back to the ‘home’ server.
have assigned tasks to individual units. By combining a wake-on-lan type feature with the
Zigbee protocol, all units can simultaneously act as coordinators, routers, and end devices.
A low-power mode can be maintained by these sensors between sampling, responding to a
wake command to route information along or to sample the environment. This design is
advantageous for our application. Where the coordinator mesh nodes may be well suited to
the larger communication nodes in our network, the sensor nodes in our network are not
particularly well suited to the always-on requirement for the Zigbee end device type. While
42
Figure 4.6: This cartoon depicts the nodal arrangement of a hybrid network. Data are shared among units
when possible, but portions of the network may share similarities with line, bus, and star networks.
it would be possible to maintain a router state for some time, multiple days of inclement
weather would negatively impact the power collected by the solar cells, and holes might
appear in the network. Therefore, using the Zigbee protocol exclusively limits the network
topology to one which eliminates the use of router nodes, commonly referred to as a “star”
network. This topology would limit the physical area covered by the network array. The
homogenization of the nodes using the Digimesh protocol eliminates these issues. One example
of the automatically generated mesh network can be seen in Figure 4.8. In this case, the
mesh generating algorithm decided it would be best to coordinate most of the sensors as end
43
Figure 4.7: An image of one of the Xbee wireless modems used in this project.
nodes connected to the unit labeled with ‘934A’ as the last four numbers in the identification
code. However, not all sensors coordinated with this like a star network. The unit with the
identifying number ending in ‘7051’ acts as a router for the unit ‘9377.’
To send data directly back to the ‘home’ server, a node in each subnet was equipped
with a long range cellular modem. The communication nodes included an extra breakout board
as described in Section 5.3.4. This breakout board utilizes the NimbeLink Skywire EVDO
cellular modem to establish communication with the home server over a cellular data network.
The Skywire device communicates using the Evolution-Data Optimied (EVDO) TIA-856 Rev.
A protocol. This telecommunications standard provides for wireless transmission of data as an
evolution of the code division multiple access (CDMA) standard [
93
]. CDMA is the technique
by which all cellular communications are based. Data from different users are multiplexed
to asynchronously access a channel and spread information across spectrum subchannels.
This technology is more suitable for sending large volumes of data, such as a collection of
data acquired by an array of sensors, than a direct single-channel communication pathway.
Additionally, by employing the EVDO standard in the sensor array, the communication nodes
44
Figure 4.8: This is a map of one of the networks used in the testing field. This is an example of a hybrid
network. The various sensor nodes all connect to a single sensor node, like in a star network. This node
connects with the communication node which, in turn, connects via cellular data to the home server.
can make use of preexisting data infrastructure to transfer the data over long distances.
With these different levels of wireless modems, a two-tiered approach can be applied to
the network. By keeping many of the components the same across tiers, the communication
nodes serve much the same function as the sensor nodes, but with increased capabilities. A
functional diagram of the communication and sensor nodes can be seen in Figure 4.9.
45
Figure 4.9: Functional diagrams for both the sensor and communication nodes in the network. Note how
the communication node has all of the same parts as the sensor node with some additions.
4.3 Data Storage and Transmission
Each tier of the sensor network is programmed to treat data differently. While all
units are designed to collect data from their own sensors, they are also designed to collect and
store the data transmitted from other nodes in the network. This creates a data redundancies
which minimizes risks to the network. If a single unit goes down, all is not lost. Even if one
of the large communication nodes is knocked off-line, the data collection is not compromised.
Each of the other units will collect and store the data on a physically retrievable SD card.
This resiliency was designed to prevent issues outside of the control of the group, such as
weather, cellular network outages, and unforeseeable damage.
46
The unprocessed sensor data and diagnostic information is collected every 15 minutes
and stored in the ring buffer (see Figure 4.10). The ring buffer also contains markers to allow
the system to continue where it left off after being reset or after a complete power failure.
About one month of data can be stored before the data must be written to the SD card.
Depending on the battery charge, the new data is written to the SD card from the ring buffer
once an hour. At this time, data is also sent to the communication nodes. If the system
charge is low, the system will wait until until the battery is significantly recharged before to
turning on the SD card for writing and transmitting the data. The SD card writing and data
transmission are performed independently with separate pointers into the ring buffer. The
power-managed SD card provides a local backup. File rotation is performed every week to
insure that the data file is kept within a reasonable size. The network routing is optimized
every hour. In the mesh setup, data packets can transfer from one node to another node as
they travel to the communication node. This increases both the range and reliability of the
system.
Figure 4.10: Data collection scheme for the sensor nodes. The data collected by the sensors are routed by
the processor into a ring buffer. These data from the buffer are periodically dumped to both the SD card
on the node as well as synced with other nodes in the network.
The communication nodes direct the self-assembly of the ad-hoc hybrid network and
collect transmitted data (see Figure 4.11). Data are received at specific times from each
sensor node and are identified by the modems 64-bit identification number. The processor
writes the data to the ring buffer and periodically transfers the data to the SD card. Again,
file rotation is performed. The SD card is written every hour or when the buffer reaches 75%
capacity. The high capacity scenario is only possible if a sensor node has delayed sending
data due to routing or power issues. Power management for the communication nodes is
47
critically important, but is mediated to some extent due to the large capacity of the solar
panel and batteries. The communication nodes have been tested with a network containing
20 sensor nodes. Currently, the number of sensors in each network can scale to 55 nodes.
Figure 4.11: Data collection scheme for a sensor communication node. The data collected by the sensors
are routed by the processor into a ring buffer. Data communicated by the mesh network is also routed into
the ring buffer. These data are stored in the SD card of the communication node until a pull request from
the home server is received.
Once an hour, the communication nodes connect to a ‘home’ server on the OSU main
campus. The scheme used to transfer the data is shown in Figure 4.12. Each sensor node
transmits the directory of the SD card to the server. At this time, it is possible for the server
to request specific files from any sensor node on the mesh network, or change performance
parameters contained on each sensor node. In a typical operation, the server looks at the
file sizes on the SD card and requests any new data as indicated by an increase in the file
size. The final line count is recorded by the server along with the new file size to ensure
that the next data transfer continues where the last one left off. The raw data are stored
in a SQL database. Each communication node has a unique database table assigned to it
for raw data. Periodically, the raw data are processed to obtain the measured values from
48
the sensors and other monitoring points. The processed data are saved in a unique table in
a SQL database. This second processing step allows all checksums to be validated, and a
variety of post-processing algorithms to be applied to the data without modifying any of the
raw data.
Figure 4.12: Scheme utilized by the communication node and server to transmit data back to OSU from
the field sites. Upon a pull request from the server, the processor on the communication node takes the
data from the SD card and sends the requested data lines through the cellular network. The server stores
these data in an SQL database, and post processing operations are executed on the stored data. Upon user
interface with the server, the data can be pulled from the database into an easy to read CSV file.
Sensor network data and node parameters are stored in SQL database which is queried
during data upload. Basic information on the sensor history is also stored in this database
and can be accessed at a website associated with the ‘home’ server. The collected data are
accessible though the SQL server. The data stored on the server can be pulled from to any
computer with the server address and login credentials. A Python script is available to select
the data the user wants to pull and store in a text file of comma separated values.
49
Chapter 5
Development of a Networked Gas Sensor Array for the Detection of CO
2
and CH
4
Concentrations
5.1 Introduction
When faced with the problem of modeling gas emissions from underground storage
facilities, a single sensor may not be adequate. This task requires monitoring of a large area
of land over long time scales. Single sensors cannot monitor such a broad area, and it is
expensive to employ a technician to monitor many sensors over time. To satisfy the needs
of this problem, A wireless sensor array was designed based on a mesh network design to
both sample the environment and coordinate remote data collection. The sensor array was
designed with sensors optimal to the conditions of the environment based on the results
discussed in Chapter 2. A set of sensors were added to a unit, a single node in the array.
Each unit was solar powered and designed to withstand the environmental conditions of a
normal field site. The each unit, or node, in the array was given the ability to communicate
with other nodes in the network. Data passed from these nodes was sent to a central server
at regular intervals to store the observations. In this manner, a remotely monitored array
was constructed to determine gas properties over a larger area than one sensor could monitor
alone.
In this chapter, the design and construction of the individual nodes in the sensor is
described. First, a simplified schematic of the network is offered and explained. Next, the
circuits employed by the sensors are described, including information about relevant breakout
50
boards. This chapter discusses the individual circuit design for power supply, processing and
memory, xbee wireless communication, cellular communication, environmental sensors, small
gas concentration sensors, larger pumped gas concentration sensors, and ground protection.
The difference between the passive and active gas sampling regimes is discussed. A specially
designed 3D printed part was produced specifically for passive sampling, and this part is
explained in detail. Finally, this chapter relates the development of the prototype boards
and movement into production of 110 units for deployment in the field.
5.2 Device Design
To best facilitate the collection and analysis of the data from remote sites, a tiered
device hierarchy was created. Most of the nodes were computationally simple, acting as end
points of the network by collecting the data. These sensor nodes require the gas sensors,
light power supply hardware, and radio for simple wireless communication to other nodes.
The middle tier of the network collects data from the sensor nodes and communicates it to
the next level. These communication nodes are computationally simple much like the sensor
nodes, but they spend more time and energy processing than the smaller sensor nodes. Thus
the communication nodes require similar gas sensors and radio, but with a more robust power
supply. The coordinated data is transmitted to the upper level via cellular phone protocols,
necessitating a cell modem. With the extra power made available to the communication
nodes, better quality sensors could be added to them, allowing verification of incoming data.
They employ a set of actively pumped sensors in addition to the passive sensors included in
the smaller nodes. The highest tier of the network would be a computational workstation
housed at the base of operations. This computer, connected to the municipal power supply,
is stationary and can afford to have any number of configurations. It acts as a receiver of the
field data. Data are sent to a web server being run by this workstation. An organized list of
requirements for the device hierarchy can be see in table 5.1.
The nodes can be generalized as a set of functional units, such as depicted in Fig-
ure 5.1. The functional units are grouped as “power”, “Arduino based processing”, and “glue
electronics.” Although the power supply capacity of each node on the tier differs, the power
supervisor and supplies required for each unit are essentially the same. The “Arduino based
processing” parts are shared between tiers as well. By producing similar board infrastructure
51
Tier and Name Device Requirements
Tier 0
Web Server
Large Storage Space
Tier 1
Communication Node
Simple gas sensors
Advanced gas sensors
Moderate power supply
Wireless communication
Cellular modem
Tier 2
Sensor Node
Simple gas sensors
Low power supply
Wireless communication
Table 5.1: Hierarchy and respective requirements of devices in the network array designed for this project
on tiers 1 and 2, parts unique to a certain tier such as the cellular modem are added on as
additional breakout parts to the central board. The “glue electronics” may be expected to
change for the more powerful sensors included with the communication nodes, but sensor
redundancy in these nodes can work with the more powerful sensors used in addition to the
simple ones. The redundancies provide the advantage of comparison of response between the
sensor types, providing calibration during the post-process.
5.2.1 Gas Sensors
Several commercially available carbon dioxide and methane sensors were selected and
evaluated for applicability to the sensor array. A full description of sensor choice can be
found in Chapter 2. Certain sensor units were chosen for specific promised abilities. Notably,
the K-30 model from CO
2
Meter was chosen and cheap, lightweight sensors which promised
satisfactory detection of carbon dioxide. The Gascard sensors sold by GHG Analytical were
an order of magnitude more costly than the other sensors, but the advertised capability made
them attractive enough to make up for the expense. The MQ-4 methane sensor was chosen
as the inexpensive methane detector; though it lacks the high sensitivity to concentration
changes which the other sensors possess, it will serve to detect large shifts in the methane
present. Due to the unique interface requirements of each sensor unit, development kits were
purchased when possible for prototyping.
52
Figure 5.1: A generalized representation of the equipment to be included in each node.
5.2.2 Power Management
The nature of the sensor array designed to cover a relatively large area of land imposes
limitations upon the power supply of the system. Since the area to be covered may include
elevation changes, anthropogenic developments, and other uncontrollable factors, it would
not be convenient to constrain the array by wired connection as the array will be adapted
53
to individual site conditions. Therefore, each unit must be designed with internal and
self-sustaining power supplies. Two different types of commercially available solar power
generation and storage units with weatherproof enclosures were selected from Tycon Power
Systems. The majority of the sensors would be housed in the Remote Pro 2.5 W Continuous
Remote Power System die cast enclosure (Figure 5.2 A) and a few sensor units with features
requiring more power were housed in the Remote Pro 15 W Continuous Remote Power System
steel enclosure (Figure 5.2 B). The 2.5 W enclosure includes a 12 V battery rated for 9 Ah of
use, a charging and distribution circuit, and a 10 W solar panel. The 15 W enclosure includes
a 12 V battery rated for 98 Ah, a charging and distribution circuit, and a 60 W solar panel
[86].
Figure 5.2: The (A) Remote Pro 2.5 W Continuous Remote Power System and (B) Remote Pro 15 W
Continuous Remote Power System–image from Tycon Power Systems’ website shows typical units.
54
During prototyping of the the devices using readily available Arduino development
boards, the power consumption of the unit was found to quickly outstrip the combined
charging and storage of power within the unit. The individual sensors were tested with
a Fluke 179 True RMS Multimeter to ascertain the current and voltage across the power
leads to verify the values provided in the specifications for each part. The calculated power
from each of these measurements is summarized in the first part of Table 5.2. The Arduino
prototype was found to use 2.7 W of power, over the continuous use rating of 2.5 W for the
small enclosure power system. By subtracting the sensor power from the total consumption of
the Arduino prototype, it was determined that the Arduino circuitry was too power intensive
for this application. Thus it became clear that a dedicated power supply circuit would need
to be designed for the deployed units.
Table 5.2: Power use of sensors examined, information based on manufacturer data sheets.
Device Type Part Power Use (mW) Used in Deployed Model?
CO
2
Sensor
K-30 SE-0018 Yes
Gascard CO
2
Yes
CH
4
Sensor
MQ-4 Yes
Gascard CH
4
Yes
DC/DC Converter
A Yes
B No
C No
D Yes
5.3 Circuit Description
Figure 5.3 shows several views of the control board. In Figure 5.4, the sensor node
control board is shown. This unit utilizes the same PCB as the control board used in the
communication node, but less of the parts are filled in. The figure is labeled to show the
location of parts on the board relating to some of the important circuits. In Figure 5.5, the
communication node control board is shown. This PCB is completely filled with components.
The labels on this figure relate to the communication board-specific circuits, and the circuits
mentioned previously in Figure 5.4 are present as well.
The sensor command module was designed in such a way as to minimize wasted power.
A selection of parts considered for use in the board were measured for current and voltage to
55
Figure 5.3: The fully built communication node command board designed for this study.
determine their power use, and these results are summarized in the second part of Table 5.2.
The control board designed for this project uses
1.7 W
during full use. To prepare for times
when the solar panel cannot provide continuous charging, a problem which can last for long
periods in the deployment area, 32% of the year is at least half-cloudy and 8% of the year
is heavily overcast [
94
]. A lower power use mode was programmed. The control board can
select a medium mode which reduces the frequency of sampling by some of the more energy
expensive sensors. This is automatically triggered by the control program if the battery
charge is determined to be at 70% charge. A hibernation mode was also programmed. This
mode disables the sensors, communication circuitry, and microprocessor, leaving only power
supply and real-time clock connected to the circuit. Below a certain battery voltage threshold,
the unit can hibernate to prevent complete power drain, and it will restart again once the
battery registers above the threshold voltage. Using only
0.1 mW
during hibernation, this
mode prevents the units from requiring manned intervention during inclement conditions.
56
Figure 5.4: This diagram outlines some of the most important circuits placed on the control board used
for the sensor nodes.
Figure 5.5: This diagram outlines important circuits on the control board used for the communication
nodes which differ from those included on the sensor nodes.
5.3.1 Power Supply
The unit is powered from the case supply circuit through 22 AWG wire connected via
terminal block. This 12 V line goes through a 1 A cartridge fuse and powers several circuits
(Figure 5.6). The 12V line sends power to a 3.3 V DC/DC converter circuit, two separate 5 V
DC/DC converter circuits, and the power supply for the Gascard (for the larger nodes only).
The 12 V goes directly into a Murata Power Solutions OKI-78SR-33/1.5-W36-C fixed output
1.5 A DC/DC converter which steps down the DC current from 12 V to 3.3 V (Figure 5.7).
57
Figure 5.6: Fuse protected 12 V terminal circuit.
Figure 5.7: 3.3 V DC/DC converter circuit.
Each of the 5 V DC/DC converter circuits begins by running the 12 V line through a
STMicroelectronics VN5016AJ-E single channel high side driver which delivers current of
a clamped voltage to components further downstream (Figure 5.8). The VN5016AJ-E is
switched by a current sense input. When this current sense is low, the current delivered by the
chip is proportional to the load current based on a known ratio. This switch is controlled by
the microcontroller based on current sense readings. The current of the voltage clamp is passed
through a Texas Instruments TLV2471IDBVR 600
µ
A 2.8 MHz rail-to-rail input/output high
58
drive operational amplifier which is operating as a closed loop non-inverting amplifier. The
high side driver delivers the power to the input of a Murata Power Solutions OKR-T/1.5-
W12-C adjustable output 1.5 A DC/DC converter. This device efficiently converts power
between 0.591 V and 6 V based on trim, and it may be turned off when a separate pin is
pulled low. The power delivered to the unit by the high side driver is used to pull the on/off
pin high with a 10K resistor. The calculated resistance required for the trim pin to achieve
5.00 V is 268 Ω. A 270 Ω resistor was determined to produce voltage within bounds of the
requirements for the components powered by this converter without requiring an expensive
non-standard resistor value. The 5 V line produced by the converter is monitored by the
ATmega 2560 chip on an A/D pin protected by resistors. An optional terminal block header
was installed to monitor the voltage being produced by the solar panel. This simple circuit is
protected by resistors and connects to one of the A/D pins on the ATmega 2560.
5.3.2 Processor and Memory
The board commands are coordinated by an Atmel ATmega 2560 lower power 8-bit
microcontroller with 64 kB flash operating on a RISC architecture (referred to as ATmega
2560 in this manuscript). The ATmega 2560 is connected to a MA-506 8MHz
±
30 ppm
crystal oscillator with equivalent length traces (Figure 5.9). Several push-button switches
with associated LEDs are connected directly to the controller to act as human interface
for simple commands such as reset, on/off, and dump log (Figure 5.10). Interface with the
RISC architecture can occur through either a TTL/RS-232 header or an in-circuit serial
programming header for programming firmware or debugging (Figure 5.11). These headers
connect directly to the ATmega 2560 and do not pass through any RS-232 type translators.
Logging is managed by the controller using timestamps from a DS3231 real-time clock
(Figure 5.12). The real time clock uses a CR-2016 coincell battery to maintain correct count
of the date. Data can be logged using two memory storage methods. The first is 24LC1026
1024 kb serial EEPROM (Figure 5.13). The second is a micro Secure Digital stable storage,
an Alps SCHD3A0100 micro SD card holder which is activated by a Vishay Si2377EDS
p-type MOSFET connected to the ATmega 2560 chip (Figure 5.14).
59
Figure 5.8: 5 V Dc/DC converter circuit
Figure 5.9: External crystal oscillator.
5.3.3 XBee Wireless Communication
Communication of the nodes with the rest of the mesh will occur by way of an XBee
wireless device. For this application, the XBee-Pro 900HP model was chosen (Figure 5.17).
60
Figure 5.10: Example of switch and associated LED.
Figure 5.11: ICSP header pinout.
Figure 5.12: DS3231 real time clock circuit.
This 900 MHz radio module is capable of data transfer rates of up to 200 kbps with impressive
range, depending on the antenna gain. The XBee communicates directly with the ATmega
2650 chip via TX/RX serial communication. Traces also directly connect with the board
to send signals telling the radio to reset and sleep. The circuit has some human interface
61
Figure 5.13: EEprom memory IC and connections.
Figure 5.14: microSD memory slot with switch circuitry.
components. One LED provides “on” status, a pushbutton switch is set to the “commission”
function of the XBee firmware, and another LED is tied to the “associate” function. The
commission button allows the user to wake a device, send node identifying codes, broadcast a
request to join a network, and restores default values depending on the number of times the
button is pressed. The associate LED provides information about the status of the network
and diagnostics which can be interpreted by the rate and number of blinks by the diode. The
device is powered by the 3.3 V power line, and the sleeping of the modem is controlled by
this same power line. When power begins flowing through the 3.3 V line from an OFF state,
an On Semiconductor 2N7002E n-type MOSFET sends a signal to the XBee to activate it. A
signal can be sent from the XBee to reset to ATmega 2560 as well. The signal is delayed
slightly by a capacitor before another n-type MOSFET sends a signal to the control board
with the reset signal.
62
5.3.4 Cellular Communication
Communication by the control nodes to the data collection site takes place over cellular
phone lines. A breakout board is connected to the control board by a ribbon patch cable. The
breakout board houses a NimbeLink Skywire EVDO cellular modem. The cellular modem
requires a stable 4 V power supply, so a dedicated power circuit was designed (Figure 5.18).
Much in the same way the other power supply circuits were constructed, this line draws
the 12 V from the battery to a VN5016AJ-E high side driver with switched current sensing
using a TLV2471IDBVR op-amp to monitor the current at this point similar to the 5 V
power supply circuits. The power is converted using a similar Murata OKR-T/1.5-W12-C
DC/DC converter with a 348 Ω resistor dialing in the output voltage required for the modem.
Monitoring of the power supply and switch is processed by the ATmega 2560. The cellular
modem is powered by the 4 V power line, but a reference voltage to the input/output lines
is also required. Since the controller uses 3.3 V, this reference voltage needs to reflect that.
The modem communicates with the ATmega 2560 by TX/RX serial port. Two input output
lines from the control board connect to n-type MOSFETs. These act as switches to send
signals to the modem for on/off and sleep operations.
Figure 5.15: This breakout board for which was attached inside the communication nodes was designed to
accommodate a NimbeLink Skywaire EVDO cellular modem.
5.3.5 Sensor Breakout Board
While the communication nodes have some more powerful sensors than the end nodes,
all types share a set of the inexpensive sensors. These parts are collected on a breakout board
referred to in schematics as the “sensor board” (Figure 5.19. Data are transfered to and
63
Figure 5.16: This is an example of the cellular modem utilized in the communication nodes. The two U.FL
connectors allow attachment of antennas for both cellular communication as well as GPS synchronization.
Figure 5.17: Schematic of XBee device as built into communication boards
from the control board by a 20 pin header connected by ribbon cable with power supply and
ground cables. Power lines in this cable use two wires each to prevent wire failure. Several
ground wires are used on the cable, positioned such that they alternate with the signal wires
on the ribbon to prevent interference between lines. The K series carbon dioxide sensor is
powered by a 5 V power line and communicates with the controller via I
2
C SDA/SDL port
64
Figure 5.18: Cellular communication breakout board schematic.
(Figure 5.21). The I
2
C is translated through an NXP PCA9517A level translating repeater
to shift the low voltage signals, which is limited by an internal resistor, up to the 5 V signal
which matches the input voltage. This part was originally intended to shift the voltage from
the 3.3 V I
2
C output from the K-series sensor. As the design progressed, the microcontroller
requirement was discovered to be 3.3 V input. Due to the timeline of the project, the sensor
boards were already printed with the level translator to convert the signal to 5 V. A separate
level converter on the main board converts the signal back down to 3.3 V (Figure 5.22). This
part of the circuit is vestigial. If one part of the device is changed in future versions, this
grandfathered circuit may prove useful once again.
The MQ-4 methane sensor is mounted on the board by connecting it to a special MQ
socket for the MQ series of gas sensors (Figure 5.25). The pin which connects to the heating
coil of the MQ-4 sensor is powered by the 5 V power line. If 5 V power is being supplied to
the sensors board, the heater will be powered on. The two other input pins on the MQ-4
sensor connect to 5 V power lines. These pins are connected after the sensor and pulled
down by a 10 kΩ resistor before connecting to an op-amp. The op-amp is configured as a
closed loop non-inverting amplifier to boost the analog signal output by the sensor. On the
mainboard, the analog signal from the methane sensor is routed to a Maxim MCP3202 12
bit analog/digital converter with serial peripheral interface (Figure 5.26). The MCP3202
splits the analog signal into a digital component with comparison to a reference clock signal
produced by the microcontroller, and it can be shutdown with a signal from the controller to
65
Figure 5.19: A picture of the sensor breakout board with encapsulating materials removed.
Figure 5.20: The K series CO
2
sensor is an optical sensor which detects gas passing through a membrane.
This is mounted on a discrete board which is mounted with standoffs on the sensor breakout board.
reduce power consumption. Digital output from the A/D converter and digital input to it
must pass through a Maxim MAX3390E low-voltage level translator. This converts the 5
V signal used by the sensors to the 3.3 V signal used by the ATmega 2560. The tri-state
outputs of this level translator minimizes the current used by the chip in the translation, an
66
Figure 5.21: CO
2
sensor with level translator on sensor breakout board.
Figure 5.22: Level translator on mainboard connected to CO
2
sensor.
important feature for power conscious design.
Two gas property sensors are connected to the board. The first is a Sensiriron SHT75
humidity and temperature sensor (Figure 5.29). This device contains the amplifier, A/D
converter, and memory necessary for operation. The device is connected to the main board
with only a few resistors such as a 4.7 kΩ pull up on the data line, a 4.7 kΩ pull down on the
clock input line, and a 220 Ω resistor on the data line to resist inrush current on the long wire
to the microcontroller. On the main board side, the clock input is switched and inverted by a
MOSFET (Figure 5.30) and the data line is adjusted to the 3.3 V input of the microcontroller
by a PCA9517A level translator (Figure 5.31). The second is a Freescale MPXA6115AC6T1
absolute, integrated pressure sensor (Figure 5.32). Since this sensor package includes the
circuitry to integrate the signal, all that is left to do on the circuit board is to boost the
67
Figure 5.23: The MQ-4 CH
4
sensor is a pluggable unit. The wire mesh prevents damage to the delicate
sensing parts, and, according to manufacturers, prevents explosions when applied to dangerous gas mixtures.
Figure 5.24: This image of the MQ-4 sensor with the metal mesh removed shows the actual sensor. A gas
sensing layer on an Al
2
O
3
substrate is connected with Au and Pt electrodes. A nichrome filament housed
within the ceramic tube heats the area during detection to improve response.
68
Figure 5.25: CH
4
sensor socket with amplifier.
Figure 5.26: Serial peripheral interface and level translator connected to CH
4
sensor.
signal with an op-amp closed loop non-inverting amplifier circuit.
5.3.6 Gascard Implementation
The larger communication nodes have an additional sensor, the Edinburgh Gascard,
for calibration purposes which has increased power requirements compared to the low power
sensors. This sensor has several dedicated circuits for its operation. The Gascard operates
under 12 V conditions, the same as is supplied by the battery, negating the need for a
DC/DC converter. The power supplied by the battery passes through a VN5016AJ-E high
69
Figure 5.27: The Sensiriron SHT75 temperature and humidity sensor consists of a small probe sticking
which sticks out from the board. The end of this probe has a comparatively thick layer of metal monolith
which acts as a heat sink to prevent erratic temperature changes which do not accurately reflect the true
average temperature of the environment.
Figure 5.28: The Freescale MPXA6115 pressure sensor detects the external pressure with a membrane
protected by a long stem tube. Though previous application of this sensor by the group have used tubing
to connect the sensor on the board to the outside of the case, the design of each sensor units allows direct
access.
side driver which is used to clamp the voltage and monitor the current output, amplified by a
TLV2471IDBVR op-amp (Figure 5.33). The output of the high side driver is fused to protect
the circuitry and connects to a terminal block header. A separate lead from the fuse, which
is protected by resistors, is used to monitor the voltage output.
70
Figure 5.29: SHT75 schematic on sensor board.
Figure 5.30: MOSFET switches used with the SHT75 clock line.
Figure 5.31: SHT75 data level translator on mainboard.
71
Figure 5.32: MPXA6115A pressure sensor and amplifier.
Figure 5.33: Gascard 4 V power supply circuitry
Unlike the membrane sensors used in the sensor nodes, the Gascard requires flow to
operate. All sensors used in the communication nodes will therefore be operating through
flow enclosures. This flow is produced by a Thomas 1410D/2.2/E/BLDC diaphragm pump.
The pump is powered by the 12 V line and uses the same fused high side driver and op amp
circuit as the Gascard (Figure 5.34). The pump is connected by a four pin header. Two of
the pins are connected by a Bourns 3361P-1-103GLF potentiometer. This Cermet trimpot
can be adjusted from 10 Ω to 10 kΩ, and it is used to adjust the speed of the pump.
72
Figure 5.34: Pump power and adjustment for communication node main boards
Data communication to and from the Gascard occurs by means of a 10 pin header which
connects the board to the sensor by ribbon cable. Communication with the microcontroller
is translated through a Texas Instruments MAXRS3222 multichannel RS-232 line driver/
receiver integrated circuit which translates the serial output of the Gascard to an asynchronous
communication protocol (Figure 5.35). The IC is configured for operation at 3.3 V, using
0.47
µ
F capacitors on the charge pump capacitors. These capacitors are larger than the
minimum required by the circuit, but the increased capacitance will reduce ripple current on
the transmitter outputs, decrease power consumption, and prevent equivalent series resistance
issues. The receive and transmit lines of the serial serial connection to the microcontroller
are connected to the receiver output and driver input on the chip, respectively, and the serial
receive and transmit coming from the gascard connect to the driver output and receive input
on the chip. A digital input/output line from the microcontroller connects to a shutdown pin
on the IC which is pulled down.
73
Figure 5.35: Serial to asynchronous communication translator assigned to Gascard
5.3.7 Ground Protection
Many components of the board are protected by capacitors to ground. This is to
prevent AC power fluctuations on DC subcircuit lines. The capacitor allows the AC signal
component to take the preferred path to ground by high frequency inductance while forcing
DC signal to still go to the intended location. Various capacitors are used for this application
in the circuit including 10
µ
F 25 V rated electrolytic capacitor with 20% tolerance, 1
µ
F 50 V
rated electrolytic capacitor with 20% tolerance, 1
µ
F 25 V rated tantalum capacitor with 10%
tolerance, 0.1 µF 25 V rated multilayer ceramic capacitor (MLCC) with 10%tolerance, 0.47
µ
F 50 V rated MLCC with 10% tolerance, and 47 pF 100 V rated MLCC with 5% tolerance.
The values are selected to filter out noise expected from that signal. Many of the values were
74
selected based on manufacturer recommendations for the associated components. Larger
capacitors were chosen to eliminate sustained voltage drops, while the smaller capacitors
were chosen to eliminate fast transient noise. A tantalum capacitor was used in applications
that required a larger capacitor with low reactance. Multiple decoupling filters are used in
series on lines where a wide range of signal noise may be encountered.
5.4 Air Sampling
Samples of the ambient air at the field site are collected by both the small sensor
nodes and the large communication nodes. To minimize the power expenditure, the small
nodes are set to passively sample the environmental conditions. The large nodes have greater
power storage capabilities, so these units sample the conditions by active sampling.
5.4.1 Passive Sampling
The passive sampling method requires the sensor components to be directly exposed
to the environment. This poses some potential problems since these components are soldered
to the circuit board. Placing an exposed circuit board into the relatively harsh conditions of
the field site invites potential issues from moisture causing corrosion, animal life interfering
with the fragile components, and damage from collisions. To mediate this problem, it was
determined that a plastic housing to isolate the sensor components from the circuit boards
and power supply electronics should be constructed. With the solar panel enclosures already
selected, the sensors would be placed in such a manner that they would have contact with the
environment through holes pre-drilled on the enclosure. These holes are oriented towards the
ground to prevent collection of rainwater and moisture. A preliminary plastic housing was
designed by the team which would mount on the carbon dioxide sensor and stick out of one
of these holes (see Figure 5.42). Expanding on this principle, a more complex plastic housing
for individual sensor groups was proposed. To prevent the high tooling expense which would
be required to construct over 100 of these relatively complex parts, a 3D printed part was
utilized.
75
5.4.2 3D Printed Parts
Printing 3D parts is a process which “prints” individual layers of polymer on top of
one another to produce a three dimensional object. The plastic housing was designed in a
program called OpenSCAD, a free 3D computer aided design (CAD) program which renders
a 3D object from a script file [95].
The part design is based on simple solids. Individual components of the design are
commented within the code. The script used to generate the model can be viewed in the
appendix under Listing D.1. A labeled image of the rendered model in Figure 5.36 illustrates
these components. The use of simple solids such as a cube and a cylinder are manipulated
with basic transformations like translation, rotation, and elongation to form the more detailed
part. The bulk of the piece is an stretched cube. A thinner stretched cube is joined on the
underside of the part to give enough clearance for the membrane on the CO
2
sensor. Two
“stovepipes” are added on the top with a cylinder joined to the original cube and a narrower
cylinder cut through these parts. The spacing of these two stovepipes corresponds to two of
the pre-drilled holes in the sensor enclosure. Four screw holes are cut with cylinders at points
which match the screw holes on the sensor circuit board. The holes have countersinks which
keep a flush surface for mounting the part on the inside of the enclosure. A hole is cut on the
edge with two cylinders to provide room for an antenna to project through a third pre-drilled
hole in the enclosure and a room to allow the a covered vent on the enclosure to have exposure
to the outside air, preventing rupture of the seal during pressure fluctuation. The pointed
overlap of these two cylinders is cut with a cube to make this edge cleaner. A notch is cut
with a cylinder to give room for a support brace nodule inside the enclosure. Areas referred
to in the diagram and script as “digs” are cut out with several stretched cylinders and cubes
to reduce the amount of material used for the part while still retaining structural integrity.
A “lower portion cutoff” was cut from the edge which would be near an inside edge of the
enclosure which was designed with a fillet. Since this cut area overlapped with a screw hole,
some more complicated cutting was performed on the model near this screw to allow room
for the fillet while still providing support for the screw. Finally, a part identified in the script
as the “OptionalBox” was added on the underside of the part which is a thin hollow cube
that acts as a containment for the parts inside by butting up against the circuit board.
The original plan was to print all of the parts on 3D printers owned and operated
76
Figure 5.36: Rendering of plastic housing with component labels used in the OpenSCAD script. Naming
convention carried over from shorthand used in code.
by Oklahoma State University. An early version of the model was printed on a MakerBot
2 using fused deposition modeling of softened polylactic acid (see Figure 5.37). The unit
specifications list layer resolution of 100 µm for the MakerBot 2.
Analysis of the printed object with a 10x magnification loupe show that the part
which was printed by the university actually appears to have a 200
µ
m layer resolution in
the XY dimension (see Figure 5.38) and in the Z dimension (see Figure 5.39), with respect
to orientation of the printed object on the build platform. A model was ordered from an
outside source, i.materialise, to compare quality. The model from i.materialise was printed
on an EOSINT P 700 by selective laser sintering of polyamide granules. The manufacturer
77
Figure 5.37: An early attempt at producing the 3D part on a Makerbot 2 3D printer.
specifications for this printer also claim a 100
µ
m layer resolution, depending on the source
material used. Analysis of the surface of the object under the same 10x loupe showed a much
more amorphous surface. Photographs were taken comparing the surface in the same areas
examined for the MakerBot 2 produced unit (see Figures 5.38 and 5.39). In these images, it
is much more difficult to distinguish defined layers, and the surface has less intense surface
deformation. The final order of the 3D printed parts was ordered from i.materalise due to
the preferable ‘cleaner’ appearance.
The part is screwed into 3
/
4” standoffs which provide clearance from the board and
sensors (see Figure 5.43). An isolating enclosure for the temperature and pressure sensors
is mounted flush against the PCB and the void is filled with an epoxy potting compound.
Enough of the potting compound is injected into this hole to cover the leads of the pressure
and temperature sensors within without overfilling and preventing these sensors to operate.
This effectively seals the sampling areas which are exposed to the environment, off from
interior of the enclosure. No potting compound is necessary for the carbon dioxide sensor,
as the extra bottom added to the cube compresses the outer edge of the membrane on that
sensor and forms a satisfactory seal (see Figure 5.44). When the combined sensor circuit
78
Figure 5.38: Flat mounting surface of part printed on MakerBot 2 on 3D printed part surfaces under 10x
magnification loupe. Primary tick marks on ruler denote 1 mm increments, and the secondary tick marks
denote 100 µm increments.
board and plastic part are attached to the enclosure, a small amount of silicone is applied to
the flat surface of the plastic part, and the piece easily slips into the correct position on the
enclosure (see Figure 5.45). Due to problems with small invertebrates making homes in the
sensor holes during the prototyping phase, a 1 mm mesh screen is glued over the holes before
the unit is deployed.
5.4.3 Active Sampling
The larger power reserves in the communication nodes allow active sampling of the
environmental conditions. A Thomas 1410D/2.2/E/BLDC diaphragm pump pulls air from
outside of the controller through a hole on the underside of the enclosure, through a 0.45
µ
m
particle filter, to the pump diaphragm. Air is pushed out from the pump to a small plastic
housing with the sensors that are used on the node sensors, then through the Gascard, and
finally vented outside the enclosure. Since it was essential to use the same sensor board in
the communication node as is used in the sensor nodes to enable direct comparison of results,
79
Figure 5.39: ‘Stovepipe’ surface of part printed on MakerBot 2 on 3D printed part surfaces under 10x
magnification loupe. Primary tick marks on ruler denote 1 mm increments, and the secondary tick marks
denote 100 µm increments.
special consideration was needed to adapt the passive sampling sensors for the flow mode in
units with active sampling. To do this, a plastic housing was designed to hold the sensor
board (see Figure 5.46).
80
Figure 5.40: Flat mounting surface of part printed on EOSINT P 700 on 3D printed part surfaces under
10x magnification loupe. Primary tick marks on ruler denote 1 mm increments, and the secondary tick marks
denote 100 µm increments.
81
Figure 5.41: ‘Stovepipe’ surface of part printed on EOSINT P 700 on 3D printed part surfaces under 10x
magnification loupe. Primary tick marks on ruler denote 1 mm increments, and the secondary tick marks
denote 100 µm increments.
Figure 5.42: Early prototype of plastic parts to isolate and protect sensor components.
82
Figure 5.43: 3D printed part attached to the standoffs on the sensor board, showing the flush interface
with the board and the carbon dioxide sensor.
Figure 5.44: 3D printed part with silicone ready to be inserted into the enclosure.
83
Figure 5.45: The custom part joins the sensor board and enclosure perfectly.
Figure 5.46: Lexan box housing sensor board for use in communication nodes. The box has been sealed
from external flow by dichloromethane solvent at the joins.
84
5.5 Prototyping
Early iterations of the sensor board were based on the Arduino interface circuitry
developed during sensor selection. This eliminated the need to develop a communications
board before the sensor board as assembly plans were developed. The Arduino setup allowed
for the sensor board to be tested for correct operation of each sensor during the development
phase and provide quality control for each of the 150 completed sensor board units. To parallel
the tests performed in laboratory setting, a series of tests were performed by constructing
prototype models to conduct tests for field reliability. The prototypes utilized the completed
sensor board, enclosure, solar power collector, and a simplified control board. The simplified
control board consisted of an Arduino Mega with a commercially available Assembled Data
Logging Shield purchased from Adafruit Industries, LLC. We chose to use Arduino Mega
since it is based on the same microcontroller designated for use in the control board. The
shield was chosen due to the ample breadboarding area, simple SD card logging, and real-time
clock. The breadboarding area of the shield was fitted with a 20 pin interface for the sensor
board and a rudimentary DC/DC power converter circuit. Power from the 12 V supply in
the enclosure was converted to the 3.3 V required by the Arduino and sensors using one of
the DC/DC power converters being investigated for use on the final model. The internal
setup of the enclosure is pictured in Figure 5.47.
A second generation prototype was developed replacing the Arduino and shield with
a hand-soldered prototype of the final control board. The updated internals of this board
can be seen in Figure 5.48. This sensor unit was deployed in a residential section southwest
of the Oklahoma State University main campus. These tests showed that the printed circuit
board functioned identically to the early Arduino prototype model.
85
Figure 5.47: Arduino Mega development board with an real-time clock and SD card installed in an enclo-
sures for testing.
86
Figure 5.48: The completed sensor node was made by replacing the prototyping Arduino Mega with the
custom circuit board designed for this project. This shot of the internals shows the detail of the wiring
between the components in the case.
87
5.6 Mass Production
The circuitry for the individual units was produced in stages. The sensor breakout
board was printed and assembled early in the first quarter of 2015, the control board for both
the communication and sensor nodes was printed and assembled in the 3
rd
quarter of 2015,
and the cellular breakout board was printed early in the fourth quarter of 2015. All printed
circuit boards and pick-and-place assembly was contracted through Advanced Circuits in
Aurora, Colorado. The sensor boards, being designed first, were the first to be constructed.
These completed sensor boards were used during the Arduino prototyping phase so device
design would employ the actual sensors. This enabled development and honing of the serial
communication with the sensors to and from the microcontroller. The control boards were
produced after several developing several prototypes and printed in a single batch. Since the
communication and sensor nodes utilize the same control board with different parts filled
in, these boards were printed in a single batch. The pick-and-place assembly of parts was
done as two separate orders. The cellular modem breakout board was only printed, due to
the small number of parts required, and the simplicity of components. These boards were
hand-soldered.
Construction of sensor nodes began in the beginning of the second quarter of 2015.
One of the sensor nodes was assembled and documented. Using this documentation, a
detailed instruction manual was produced. These instructions are included in Appendix E.
The remaining nodes to be constructed, approximately 110 units, were assembled without
the control boards by collaborators in the Oklahoma State University Civil Engineering
department.
Due to small delays in the prototyping phase of the control board, the partially
completed units were stored in the lab used for construction for over two months. Once the
control boards were received, plans were made to finish the construction. It was discovered
that during this time, all of the batteries had discharged. Multiple chargers were purchased
to expedite the correction of this issue. However, the batteries had discharged beyond the
chargeable threshold (discussed in detail in Chapter 3), and would require much more expensive
charging equipment which neither the Chemistry nor the Civil Engineering department
possessed. It was determined to be less costly to purchase all new batteries and recycle the
88
Figure 5.49: A delivery of control boards from the assemblers.
old. Installation of the batteries occurred simultaneously with the installation of the control
boards. This final assembly phase was completed in waves of ten units at a time. After
the engineers assembled the nodes, the devices were given a quality control inspection and
programmed. The programmed sensor nodes were taken to the a proving ground north of
campus and mounted on posts in sets of 10 sensor units. Each of these sensor cohorts were
networked with one of the prototype communication nodes and allowed to collect data for
several weeks. The data produced by the units were analyzed, and any devices producing
unsatisfactory results was taken for repair. Analysis of these data are found in Chapter 6.
The final cohort was deemed satisfactory in the 1
st
quarter of 2016.
Construction of the communication nodes began near the end of the 1
st
quarter in
2015. Clear Lexan boxes were ordered to built by the Civil Engineering mechanical shop
to house the sensor breakout board for active sampling (see Section 5.4.3). Circuit boards
were mounted to an orange back plate secured within the enclosure (Figure 5.50a). Only the
top half of the plate was practical for mounting circuit boards, as the large batteries inside
took up half of the volume of the enclosure. Vinyl tubing was used to plumb the unit. Input
and output ports for gas were stuck through the bottom of the enclosure to pull gas from a
89
location with a similar orientation to the passive interfaces on the sensor nodes. The ports
were placed on opposite ends of the bottom side of the enclosure to minimize recycling of gas
which had already been analyzed. Only three of these large nodes were built at first as the
primary goal at this point was construction and testing of the sensor nodes. Originally, the
Lexan boxes were designed to easily disassemble in case modifications needed to be made on
the sensor breakout board. However, flow testing gas through the box showed that 36.2% of
the mass flow of gas through the entire system was being lost. All of the joining edges of
the clear plastic on the box, including the edge meant to be disassembled with screws, was
sealed using dichloromethane in addition to silicone seals around the ribbon cable connection.
Construction on the rest of the communication nodes resumed in late 2015, finishing in the
1
st
quarter of 2016. Each communication node was tested in the field, much like the sensor
nodes, for quality assurance of the units.
90
(a) The layout of components within the large
communication nodes. The batteries take up
most of the space inside the enclosure, so com-
ponents were arranged to be bolted to the top
of the back plate.
(b) Behind the batteries, tubing to direct the air
for the active sampling is laid. Incoming air is
passed through a filter and pumped through the
clear box containing the sensor breakout board
and the gas card sequentially before exiting the
unit.
5.7 Conclusions
The circuits described in this chapter come together to produce an effective set of
sensing units. The K-30 and MQ-4 gas sensors described in Chapter 2 have been applied to all
units in the network. For added accuracy for methane detection, a Gascard sensor has been
added to the Tier 1 Communication Nodes. This allows for the signals detected by the MQ-4
91
sensors on the network to be compared against a more reliable sensor to quantify potential
concentration spikes. The units are powered by switched 12V to 5V DC/DC converters
which draw power from the enclosure’s solar charged batteries. The AVR processor issues
commands to collect and store the data on the EEPROM memory and SD card backup. The
processor also handles power management by shutting down power intensive tasks such as
the MQ-4’s heater during periods of low battery voltage. Communications between units
are carried out by an XBee wireless radio with a limited range. Communication nodes also
incorporate a cellular modem breakout board which allows for communication of data across
3G cellular networks to a local server. Individual sensors for gas concentration, pressure,
temperature, and humidity measurements are incorporated onto a sensor breakout board
which is mounted with access to the atmosphere. The atmospheric interface is a passive
sampling regime using a 3D printed part to protect delicate electronics for the low level sensor
nodes. The atmospheric sampling in communication nodes is an actively pumped regime
to allow for the use of the ’flow-only’ Gascard, and a special housing is used for the sensor
breakout board to allow adaptation to this regime.
These individual parts working in concert allow for low-power operation, independent
of human intervention for long periods of time. The sensing devices are separated by sampling
regime and location in the units, while wired to the storage architecture to collect the data.
The wireless networking capability of the units allows for the transfer and coordination of
data as outlined in Chapter 4.
Devices have been produced in quantities of
>
100 units. The increased production scale
allows for these units to be produced cheaply. The circuit boards and sensor infrastructure
were assembled and programmed in batches during the 1
st
quarter of 2016. The testing of
these units will be described in subsequent chapters.
92
Chapter 6
Results of Prototyping Tests and Long-term Tests of a Networked Sensor Array at a Proving
Ground on the OSU Campus
6.1 Introduction
During the course of unit construction and further deployment, a single group of
sensors was maintained at the proving ground on to the North of the Oklahoma State
University campus. As the sensors progressed through the design phase and through several
changes in programming, these devices collected data from the environment. These units
monitored the area without any controlled gas test, and this data was used as a control set.
6.2 Preliminary Deployment and Unscheduled Mechanical Stress Testing
Originally, four of the prototype units were mounted on steel T-posts and deployed in
a 4 yd. square North of the Oklahoma State University main campus, see Figure 6.1. The
units have demonstrated resilience against extreme weather and wildlife in the area. Data
from the unit were collected on SD cards at intervals to monitor the sensors for power failures
and abnormal behavior of the sensors. The prototypes in question were simple sensor boards
with Arduino Mega data collection equipment described in Chapter 5.
6.2.1 Detected Events
The relatively heavy power consumption of the prototype units forced the sensors
during certain periods with heavy cloud cover as is apparent in Figure 6.2. After a sufficient
93
Figure 6.1: Field testing in a 4 yard square north of the OSU campus.
voltage was detected in the battery, the unit would power back on and resume collecting
data. An interesting spike in the concentration of carbon dioxide can be observed on the
night of June 3
rd
. During a routine check of the data by manually removing the SD cards
due to lack of wireless capabilities in the prototype models on the morning of June 4
th
, a
pile of deer droppings was noticed and logged in front of Sensor 0004. It is speculated that
the concentration spike is due to the presence of grazing wildlife very near the units. This
suggests that the devices are sensitive respiration of local wildlife.
In Figure 6.3, there is some overlap in detection periods by the two sensors. This
illustrates one of the strengths of a distributed array. The methane sensors used in this
test were uncalibrated, and values represent an arbitrary level of methane which is assumed
to be constant in the air for sensors spaced so close to one another. While the methane
results showed the large amount of error expected, the team was pleasantly surprised to see
that the average of all detected values and the linear best fit line were so similar in each
unit, indicating that the sensor will still be valuable for assessing methane concentration as
averaged over a moderate amount of time.
94
5 / 1 0 / 2 0 1 5
5 / 1 7 / 2 0 1 5
5 / 2 4 / 2 0 1 5
5 / 3 1 / 2 0 1 5
6 / 7 / 2 0 1 5
6 / 1 4 / 2 0 1 5
6 / 2 1 / 2 0 1 5
6 / 2 8 / 2 0 1 5
4 0 0
6 0 0
8 0 0
1 0 0 0
C o n c e n t r a t i o n ( p p m )
D a t e
Figure 6.2: Carbon dioxide detection by a prototype unit in the field. Gaps represent power outages. The
large concentration spike on June 3
rd
is likely the detection of local fauna grazing near the sensors.
6.2.2 Storm Damage
During a heavy storm in the early morning hours of September 11
th
, two prototype
sensors incurred damage. The storm included heavy rains and wind gusts of up to 70 miles
per hour [
96
]. At the field site, a large piece of metal debris (the roof of a nearby shed) struck
two of the sensor units and caused heavy damage to the units (see Figure 6.4). Two units
were unharmed, and the two other units were struck and rendered inoperable in the field.
The damaged prototypes were returned to the lab for autopsy and analysis.
The first struck unit, designated UNIT 0003, suffered severe impact damage. The
damage to this unit was incurred primarily on the solar panel mounting hardware. Figures 6.5
and 6.6 show detail of the damage. During the impact, the sensor was lifted up, pulling the
95
5 / 1 0 / 2 0 1 5
5 / 1 7 / 2 0 1 5
5 / 2 4 / 2 0 1 5
5 / 3 1 / 2 0 1 5
6 / 7 / 2 0 1 5
6 / 1 4 / 2 0 1 5
6 / 2 1 / 2 0 1 5
6 / 2 8 / 2 0 1 5
0
5 0 0
1 0 0 0
R e s p o n s e ( A . U . )
D a t e
U n i t 0 0 0 2 m e t h a n e d a t a
U n i t 0 0 0 4 m e t h a n e d a t a
L i n e a r F i t o f U n i t 0 0 0 2 d a t a
L i n e a r F i t o f U n i t 0 0 0 4 d a t a
Figure 6.3: Methane detection by two prototype units. The linear fit shown indicates that the averaged
methane value detected by the sensors was very stable.
topmost mounting hardware off of the T-post. Curved scratches on the backside of the sensor
showed that it dangled for some time by a single mounting part until it was pulled off entirely
either by the gusts of the storm or the lever force of the heavy unit being held by a single
bracket. The solar panel was rendered inoperable by this damage, preventing recharge of the
battery. While this damage appears catastrophic to the sensor, testing of the sensors and
prototype board inside showed no damage.
The second struck unit, designated UNIT 0004, sustained less superficial damage. The
mounting bracket was loosened by the impact, and the unit slipped down the T-post directly
to the ground. This drop damaged antenna, as show in Figure 6.7, requiring it to be replaced.
The sensor was not working when it was retrieved. Upon checking the internals, it became
96
Figure 6.4: Severe damage to prototype sensors at test site after debris strike (seen in background) during
gusty weather.
clear that during the event, one of the power supply wires had become dislodged. When this
was plugged in, the unit started up with no issues.
While damage incurred by the sensors during the storm caused device failure, it
shows the resilience of the manufactured designs. It is not expected that sensors deployed
to field sites will regularly undergo impacts by heavy metal objects. The sensors were more
than capable of handling “normal” extreme weather events. The devices showed no water
damage, and none of the important components were impaired. Furthermore, the unlikely
impact showed that the devices were resistant to physical damage, as the enclosures properly
protected the delicate internal circuitry.
97
Figure 6.5: Damage to UNIT 0003: This image shows the steel hinge holding the solar panel on the Tycon
enclosure has been sheared away by stress during the impact.
98
Figure 6.6: Damage to UNIT 0003: This image shows the bracket which supports the solar panel on the
Tycon enclosure has been ripped by damage incurred during the impact.
99
Figure 6.7: Damage to UNIT 0004: This image shows the antenna which snapped after the impact. The
mud flecks on the downward facing side of the sensor confirms that this sensor was dropped straight to the
ground.
100
6.3 Comparing Collected Data from Long-Term Array against Accepted Weather
Report Data
As the mass production of units was completed, a set of 10 sensors was deployed in
the same proving ground area as the previous 4 prototype sensors. This new network was
dubbed the Long-Term Study array. The units deployed at this site were the first to be
completed. The goal of this network is to determine a baseline of the local concentration
of carbon dioxide and methane. This site was chosen due to the low number of expected
concentration change events due to the relative remoteness of the north campus site and
the lack of methane leaking ground sites nearby. Collection of the data from this network
is currently ongoing. The data were collected an analyzed during the project to report the
efficacy of the network for quarterly reports to the funding agency. It is important to confirm
that the devices designed during this project were capable of producing the data which was
claimed by our group.
One method of analyzing the control data is comparison with known figures. Collected
temperature, humidity, and pressure data were compared against data reported from a local
weather station [
97
]. Data collected from an archive of reports from the KOKSTILL4 was
collected and the values of the maximum, minimum, and average reported temperature,
humidity, and pressure were extracted from the data set. These values were plotted as three
lines over plots of the individual detected values of all of the sensors. The temperature
and humidity data, shown in Figure 6.8 and Figure 6.9 respectively, are very similar to the
weather station data. The sensors show a very regular cycling of these values, corresponding
to the day-night cycle. The peaks and troughs of this cycle match closely with the minimum
and maximum weather station data. The plot of pressure data (Figure 6.10) differs, in that
the weather station data and the test data seem to be offset. The altitude of the KOKSTILL4
station (935 ft. above sea-level) does not differ enough from the test site (944 ft. above
sea-level) to suggest that there would be a significant difference in pressure due to this. It
is possible that there is some minor variation between the pressure in the two locations,
approximately 2 miles apart. The sensor used by the weather station may also be inaccurate,
there is no published calibration data for this station. It is also possible that the pressure
sensors used by the sensor nodes are incorrect. Due to the large number of sensors involved,
101
this is improbable.
1 / 1 / 2 0 1 6 1 / 8 / 2 0 1 6 1 / 1 5 / 2 0 1 6 1 / 2 2 / 2 0 1 6 1 / 2 9 / 2 0 1 6 2 / 5 / 2 0 1 6
- 2 0
- 1 0
0
1 0
2 0
3 0
4 0
T e m p e r a t u r e (
o
C )
Figure 6.8: Temperature data collected from test sensors tracks with the reported weather data. Diurnal
cycling is apparent. Weather data is from KOKSTILL4 weather station [97]. The dashed line and dotted
line are the maximum and minimum value observed for each date, and the black line is the average value
observed for that date.
As there is no local reporting agency for gas concentrations, validation of the values
reported by the sensors is not as simple. Instead, the data were considered good if those
reported by each sensor tracked well with the other sensors in the network. Data were
collected from the sensors in the long-term study array and compared against the others in
the same network. By visual analysis, shown in Figures 6.11 and 6.12 it can be seen that the
reported values tracked well among the units. There is some variation between the units, as
the sensors were uncalibrated. Yet, the diel cycle of concentrations track well between units.
The average of the reported concentration of each gas is reasonable for the site location,
102
1 / 1 / 2 0 1 6 1 / 8 / 2 0 1 6 1 / 1 5 / 2 0 1 6 1 / 2 2 / 2 0 1 6 1 / 2 9 / 2 0 1 6 2 / 5 / 2 0 1 6
0
2 0
4 0
6 0
8 0
1 0 0
R e l a t i v e H u m i d i t y ( % )
Figure 6.9: Humidity data collected from test sensors tracks with the reported weather data. Diurnal
cycling is apparent. Weather data is from KOKSTILL4 weather station [97]. The dashed line and dotted
line are the maximum and minimum value observed for each date, and the black line is the average value
observed for that date.
based on reported background concentrations discussed in Chapter 1. The results from this
study suggest that the use of a networked array of sensors can produce an acceptable single
value for the local gas concentration by the law of averages.
103
1 / 1 / 2 0 1 6 1 / 8 / 2 0 1 6 1 / 1 5 / 2 0 1 6 1 / 2 2 / 2 0 1 6 1 / 2 9 / 2 0 1 6 2 / 5 / 2 0 1 6
0
7 0 0
7 2 5
7 5 0
7 7 5
P r e s s u r e ( t o r r )
Figure 6.10: Pressure data collected from the test sensors is acceptably precise yet consistently lower than
the reported weather data. This may indicate deviation of the sensors from the true value, local variation
in pressure, or an inaccurate report from the weather station. Weather data is from KOKSTILL4 weather
station [97]. The dashed line and dotted line are the maximum and minimum value observed for each date,
and the black line is the average value observed for that date.
104
1 / 1 / 2 0 1 6 1 / 8 / 2 0 1 6 1 / 1 5 / 2 0 1 6 1 / 2 2 / 2 0 1 6 1 / 2 9 / 2 0 1 6 2 / 5 / 2 0 1 6
0
3 0 0
3 5 0
4 0 0
4 5 0
5 0 0
5 5 0
6 0 0
C O
2
( p p m )
Figure 6.11: This plot shows carbon dioxide the data collected by the sensors in the long-term study
network. The values reported by individual sensors vary somewhat (shown in gray), but they all follow
similar patterns during the day-night cycle. The black trace shows the average value reported by all sensors
in the network.
105
1 / 1 / 2 0 1 6 1 / 8 / 2 0 1 6 1 / 1 5 / 2 0 1 6 1 / 2 2 / 2 0 1 6 1 / 2 9 / 2 0 1 6 2 / 5 / 2 0 1 6
0
1 0 0 0
2 0 0 0
3 0 0 0
4 0 0 0
C H
4
( c o u n t s )
Figure 6.12: This plot shows methane the data collected by the sensors in the long-term study network.
The values reported by individual sensors vary somewhat (shown in gray), but they all follow similar patterns
during the day-night cycle. The black trace shows the average value reported by all sensors in the network.
106
6.4 Conclusions
The scale up of the proving ground site on the north side of the Oklahoma State
University Stillwater campus from prototyping site to long-term study site shows valuable
information about the capabilities of the sensor network. In one fortuitous observation,
devices were shown to be able to detect small local concentration changes due to outside
variables such as local wildlife. This suggests that the sensors are sensitive to small, local
changes in gas concentration. Furthermore, it suggests that the time-scale on which the
sensors are programmed to operate is sufficient to detect short term events. In another
serendipitous experimental incident, the sensors were also shown to be capable of withstanding
dangerous weather. Sensors struck by debris during extreme weather conditions were shown
to have endured very little serious damage. Issues affected by this event were simple to repair.
If similar incidents are encountered in the future, it is reasonable to assume that damage
will be field-repairable. Finally, the long-term study can provide a reasonable baseline for
the local area, and the data are shown to track well as a group. Application of accepted
data from local weather stations allowed for facile verification of the device capabilities. All
sensors deployed in this group were shown to produce data reasonably close to these accepted
values. The results from this proving ground lend credence to the successful deployment of
the sensors at other test sites.
107
Chapter 7
Results of Data Collected Tests from a Networked Sensor Array at the OSU Unmanned
Airfield
7.1 Introduction
A second testing site was selected at the Oklahoma State University Mechanical and
Aerospace Engineering’s Unmanned Aircraft Systems (UAS) Airfield. This site, located
approximately 16 miles east of the main Stillwater campus, is one of the few locations owned
by a university which permits the testing of unmanned aerial vehicles. The 15 acre tract of
land includes a garage, control facility, and polymer mat runway.
The UAS site was chosen for sensor deployment as a source of collaboration between
the Chemistry and Engineering Departments on the gas sensing project. A network of sensors
developed for this project would be deployed at the site in conjunction with flights of aerial
sensors. One of these sensors can be seen in Figure 7.1. The goal of the collaborative project
is to simultaneously collect data from the sensors at ground level while a UAV mounted
sensor makes a controlled pass of the field from the air. In this way, concentration flux can
be modeled in three dimensions, rather than the two dimensions provided by the current
network setup. The site, which is managed by the University, is convenient for release testing
of analyte gases. To properly conduct these tests, it is necessary to establish background
readings for the site.
108
Figure 7.1: This image depicts one of the Unmanned Aerial Vehicles being researched at the UAS site.
This unit includes a carbon dioxide and methane flow sensor mounted in the fuselage. Collaborative efforts
are underway to compare the data detected by these aerial units to the units deployed in the sensor array.
7.2 Sensor Network
The sensor network was deployed along the fence line of the site. Metal T-posts were
erected in regular intervals along this path. A map of the site and the locations of the sensors
can be seen in Figure 7.2. Unlike the network discussed in Chapter 6 which consisted of
a single network of sensors, this deployment includes multiple subnets. In Figure 7.2, the
individual subnets are color coded. In this way, the network support for multiple subnets
could be verified.
7.3 Data Analysis
Data were successfully collected by the sensors in the network, and the data were
processed and pulled from the central home server on the OSU main campus. The data
were collected over a period of approximately 9 months, and the site is still collecting data
as of this publication. During this period, there were variations in pressure, temperature,
109
Figure 7.2: This Google Maps satellite image of the UAS airfield depicts the approximate locations of
sensors at the site. The colored groups depict the individual subnets in the sensor network.
and humidity, as can be expected of an outdoor testing period of this time scale. One of
the primary variations which can be quantified is the diel cycle. The data from each sensor
were collected and sorted by time of day. A figure representing the range of values reported
for these time points can be seen in Figure 7.3. The data from the carbon dioxide sensor
shows that the concentration seems to be highest in the early morning hours, and decreases
during the day. The opposite is the case for methane, with peak concentration occurring in
the middle of the day. This cycle matches with reports on diel concentration flux by various
authors and in various environments [
98
,
99
,
100
,
101
]. The generally accepted cause of these
variations is the flux between the ground and flora with the atmosphere. Carbon dioxide
levels increase during the night as plants cease to photosynthesize. Methane levels increase
during the day as heat from the sun generates flux of gas from that which is stored in the
topsoil. Changes in relative humidity also play a role in this cycle by reducing the partial
110
pressure of carbon dioxide and methane components with respect to the partial pressure of
the water vapor.
1.5e-05
1.6e-05
1.7e-05
1.8e-05
1.9e-05
2e-05
2.1e-05
2.2e-05
2.3e-05
2.4e-05
2.5e-05
0 4 8 12 16 20 24
Methane Conductivity (S)
360
370
380
390
400
410
420
430
Carbon Dioxide Conc. (ppm)
Figure 7.3: This plot shows the range of values recorded during each daily cycle at set time points. The
plot shows the concentration changes in the typical day/night cycle at the site.
7.3.1 Carbon Dioxide
When the carbon dioxide concentration traces are stacked in a plot, as is depicted
in Figure 7.4, the features of the plots line up nicely. This suggests that the sensors are
all functioning correctly. For any large concentration change events, we would expect to
see multiple sensors peaks. Peaks which appear in the reported data of a single unit are
likely attributable to very localized environmental changes, such as the presence of an animal
actively engaging in respiration. One trace has appears to show considerable deviation from
the other traces. This unit is suspected to have a malfunctioning multiplier, which increases
the intensity of the peaks and valleys of the detected concentrations.
When all of the data collected from the carbon dioxide sensors are averaged, the trace
depicted in Figure 7.5 is generated. This plot shows all of the same characteristics as before
111
Figure 7.4: Data collected from the carbon dioxide sensors has been stacked in this graph to show the
coincidence of peaks in the concentration.
including the diel cycles and any events detected by multiple sensors. From this averaged
dataset, we can pull several pieces of information. By plotting all of the values detected by
the sensors in single parts per million bins, we can plot a histogram of the values reported by
the sensor, as depicted in Figure 7.6. A histogram was constructed of 100 bins with a size
of 1.72 ppm. The data plotted on this histogram can be treated as a normal distribution.
The peak average is 414 ppm. This is appreciably close to the mean global concentration of
carbon dioxide at sea level. There is minimal skew with slight tailing to higher concentrations,
possibly due to peak events. The kurtosis metric shows that the normal distribution is
weighted more heavily toward the mean.
7.3.2 Methane
When the methane concentration traces are stacked in a plot, as is depicted in
Figure 7.7, the features of the plots line up nicely. This suggests that the sensors are all
functioning correctly. For any large concentration change events, we would expect to see
multiple sensors peaks. Peaks which appear in the reported data of a single unit are likely
attributable to very localized environmental changes, such as the presence of an animal
actively engaging in respiration.
When all of the data collected from the methane sensors are averaged, the trace
depicted in Figure 7.8 is generated. This plot shows all of the same characteristics as before
112
Figure 7.5: The traces from the UAS airfield have been averaged in this graph to show a single line which
depicts the average concentration of carbon dioxide at the site.
Figure 7.6: The data produced by the carbon dioxide sensors shows a Guassian distribution.
113
Figure 7.7: Data collected from the methane sensors has been stacked in this graph to show the coincidence
of peaks in the concentration.
including the diel cycles and any events detected by multiple sensors. From this averaged
dataset, we can pull several pieces of information. By plotting all of the values detected by
the sensors in
10 S
bins, we can plot a histogram of the values reported by the sensor, as
depicted in Figure 7.9. A histogram was constructed of 100 bins with a size of
4.38 × 10
−7
S
.
The data plotted on this histogram can be treated as a normal distribution. The peak average
is near
1.98 × 10
−5
S
. Since these sensors have not been accurately ascribed a calibration
curve, we can only take note of the shape of the normal distribution. The peak is Gaussian
with tailing to higher gas concentrations. This suggests that while most of the reported values
from the sensors fall within a small window, there are several events which cause this skew.
The kurtosis metric shows that the normal distribution is weighted more heavily toward the
mean.
7.3.3 Peak Events
By applying some statistical analysis to the average of all sensors on the network
for both carbon dioxide and methane, we can determine when peak concentrations were
detected. These ‘Events’ were selected by determining if a value was greater than two
standard deviations above the average. This selection method removes the natural variation
in the diel cycle from generating a false positive. If we plot these events and their respective
concentrations against time, we can look for patterns, this plot is shown in Figure 7.10. Some
114
Figure 7.8: The traces from the UAS airfield have been averaged in this graph to show a single line which
depicts the average concentration of methane at the site.
Figure 7.9: The data produced by the methane sensors shows a Guassian distribution.
115
of these detected events, such as those early in the reported data, can be ruled out. Those
peaks are likely due to human proximity to the sensors during transport and deployment.
Few of the selected events occur at similar times. It would be a simple matter of analysis to
say that these events are unrelated and occur during the peak times for each respective gas.
No causal link to any source has been established for these events.
2e-06
4e-06
6e-06
8e-06
1e-05
1.2e-05
1.4e-05
1.6e-05
08/01/16 09/01/16 10/01/16 11/01/16 12/01/16 01/01/17 02/01/17
Methane Conductivity (S)
0
200
400
600
800
1000
1200
1400
1600
Carbon Dioxide Concentration(ppm)
Figure 7.10: This plot shows the recorded events at the UAS site. Events were selected as points which
were 2 standard deviations higher than the average at that time of day.
7.4 Conclusions
By setting up a field site near Stillwater, OK at the UAS airfield, the project members
have greater access to the equipment to perform experiments. In establishing a baseline for
these future experiments, we can see that the sensor networks report both carbon dioxide and
methane concentration within expected diel cycling regimes. The carbon dioxide concentration
at this site is higher than expected compared to the global average. By analyzing the average
concentration detected by the sensors, we have been able to single out peak concentration
116
events which occurred during the data collection period. These events do not appear to be
correlated the companion analyte gas.
117
Chapter 8
Results of Data Collected from a a Networked Sensor Array near a Gas Injection Well Field
Site near Farnsworth, TX
8.1 Introduction
The ultimate test of the networked sensor array was to deploy it at a real injection
well site. A location was selected in the panhandle of Texas near the town of Farnsworth. The
site has been well characterized by geologists, and is managed by the Southwest REgional
Partnership, a carbon dioxide sequestration consortium [
102
,
103
]. A collaboration was formed
between faculty who have had previous success in deploying new methods to characterize
carbon sequestration seepage [
104
,
105
] and faculty who have used similar technology for
networked chemical sensor design [
106
,
107
,
108
] on this project, funded by the Department
of Energy.
The site is above a pilot carbon dioxide injection well in the Morrow geological
formation. The site has a large salt dome structure, and it has been noted as a possible
storage site for carbon dioxide. The current goal of the pilot well is to use the compressed
carbon dioxide supercritical fluid to elicit the remaining valuable hydrocarbons which previous
wells have failed to completely extract. Above these mineral formations are large acreages of
farmland. At the specific site the sensors were deployed, both cotton and corn were grown
during the course of this experiment.
118
8.2 Sensor Network
In the original grant application which dictated the path of this project, the sensors
were to be deployed in a grid on the site. A map of this site is shown in Figure 8.1. The red
triangles depict the proposed location of sensors on the site, and the blue circles represent the
location of wellheads at the site. The size of these circles represents the carbon dioxide flux
from the soil in previous subsurface testing. The red line segments on the topographical map
shows a truncated outline of the Farnsworth Oil Unit, which is the region that is expected to
be effected by the injection wells.
Figure 8.1: A map of a proposed grid of sensors on part of the Farnsworth Oil Unit. Red diamonds indicate
proposed locations for the sensors in the array, and the blue circles indicate wellheads where the size of the
circle is indicative of the CO
2
soil flux relative to the others. This map was produced using GIS topological
maps of the area and an outline of the Farnsworth Oil Unit.
Upon investigation of the site, it became apparent that a full grid was not a reasonable
layout. As the topsoil is actively used in center pivot irrigation farming. Deploying sensors
within these circles would inhibit agricultural use, and create problems. Additionally, the
logistics of transporting over 100 sensors to a site several hours from the Stillwater campus
with limited manpower became an issue. Due to these restrictions, the sensors were mounted
on existing wooden utility poles at the site. This limited the number and location of the
119
possible sites. Rather than the 100+ sensors planned for, a total of 33 units were deployed
to the site in 3 subnets. The location of these sensors and networks is shown in Figure 8.2.
The chosen poles are on access roads surrounding the primary injection well, as noted in the
figure.
Figure 8.2: This Google Maps satellite image of the Farnsworth, TX site depicts the approximate locations
of sensors at the site. The colored groups depict the individual subnets in the sensor network. Sensor
locations were determined in part by the placement of existing power poles.
To facilitate installation on utility poles, a different mounting method from the existing
T-post configuration needed to be designed. A piece of U-channel material (both aluminum
and high density polyethylene were used) was fitted with holes to mount the sensor with bolts
on the broad part of the channel and two slots on each arm of the channel. These slots allow
pieces of stainless steel strapping to be fit around the pole and secure the sensor. Workers on
120
ladders could quickly mount these units with the strapping, as shown in Figure 8.3.
Figure 8.3: Workers installing the sensors on power poles at the Farnsworth, TX site. The nodes, equipped
with special mounting brackets, were strapped to the poles with steel strapping. At the request of the site
maintainers, great care was taken to not damage the poles.
The first network (blue in Figure 8.2) was deployed in April 2016, and the other two
subnets were deployed in August of the same year. The units were allowed to collect data
until February 2017, when the site was decommissioned.
8.3 Data Analysis
Data were successfully collected by the sensors in the network, and the data were
processed and pulled from the central home server on the OSU main campus. Due to power
communication errors in one of the networks (an antenna bent over), some data were not
reported back to the home server. These periods of missing data have been recovered from
121
the sensors, as all of the data collected by a network is backed up in the SD card of the nodes.
The data were collected over a period of approximately about a year before the sensors at the
site were decommissioned. During this period, there were variations in pressure, temperature,
and humidity, as can be expected of an outdoor testing period of this time scale. One of
the primary variations which can be quantified is the diel cycle. The data from each sensor
were collected and sorted by time of day. A figure representing the range of values reported
for these time points can be seen in Figure 8.4. The data from the carbon dioxide sensor
shows that the concentration seems to be highest in the early morning hours, and decreases
during the day. The opposite is the case for methane, with peak concentration occurring in
the middle of the day. This cycle matches with reports on diel concentration flux by various
authors and in various environments [
98
,
99
,
100
,
101
]. The generally accepted cause of these
variations is the flux between the ground and flora with the atmosphere. Carbon dioxide
levels increase during the night as plants cease to photosynthesize. Methane levels increase
during the day as heat from the sun generates flux of gas from that which is stored in the
topsoil. Changes in relative humidity also play a role in this cycle by reducing the partial
pressure of carbon dioxide and methane components with respect to the partial pressure of
the water vapor.
8.3.1 Carbon Dioxide
When the carbon dioxide concentration traces are stacked in a plot, as is depicted
in Figure 8.5, the features of the plots line up nicely. This suggests that the sensors are
all functioning correctly. For any large concentration change events, we would expect to
see multiple sensors peaks. Peaks which appear in the reported data of a single unit are
likely attributable to very localized environmental changes, such as the presence of an animal
actively engaging in respiration.
When all of the data collected from the carbon dioxide sensors are averaged, the trace
depicted in Figure 8.6 is generated. This plot shows all of the same characteristics as before
including the diel cycles and any events detected by multiple sensors. From this averaged
dataset, we can pull several pieces of information. By plotting all of the values detected by
the sensors in single parts per million bins, we can plot a histogram of the values reported by
the sensor, as depicted in Figure 8.7. A histogram was constructed of 100 bins with a size
122
1.7e-05
1.8e-05
1.9e-05
2e-05
2.1e-05
2.2e-05
2.3e-05
2.4e-05
2.5e-05
0 4 8 12 16 20 24
Methane Conductivity (S)
380
385
390
395
400
405
410
415
420
425
430
Carbon Dioxide Conc. (ppm)
Figure 8.4: This plot shows the range of values recorded during each daily cycle at set time points. The
plot shows the concentration changes in the typical day/night cycle at the site.
Figure 8.5: Data collected from the carbon dioxide sensors has been stacked in this graph to show the
coincidence of peaks in the concentration. Gaps represent communication errors with a single sensor subnet.
of 2.02 ppm. The data plotted on this histogram can be treated as a normal distribution.
The peak average is 478 ppm. This is appreciably close to the mean global concentration of
carbon dioxide at sea level (as was discussed in Chapter 2). The kurtosis metric shows that
123
the normal distribution is weighted more heavily toward the mean. The Gaussian distribution
shows long tail of values which were reported higher than the mean. These peak values will
be discussed in more detail in Section 8.3.3.
Figure 8.6: The traces from the Farnsworth, TX site have been averaged in this graph to show a single line
which depicts the average concentration of carbon dioxide at the site.
8.3.2 Methane
When the methane concentration traces are stacked in a plot, as is depicted in
Figure 8.8, the features of the plots line up nicely. This suggests that the sensors are all
functioning correctly. For any large concentration change events, we would expect to see
multiple sensors peaks. Peaks which appear in the reported data of a single unit are likely
attributable to very localized environmental changes, such as the presence of an animal
actively engaging in respiration.
When all of the data collected from the methane sensors are averaged, the trace
depicted in Figure 8.9 is generated. This plot shows all of the same characteristics as before
including the diel cycles and any events detected by multiple sensors. From this averaged
dataset, we can pull several pieces of information. By plotting all of the values detected by the
sensors in
10 S
bins, we can plot a histogram of the values reported by the sensor, as depicted
in Figure 8.10. A histogram was constructed of 100 bins with a size of
5.28 × 10
−7
S
. The
data plotted on this histogram can be treated as a normal distribution. The peak average is
near
3.58 × 10
−5
S
. Since these sensors have not been accurately ascribed a calibration curve,
124
Figure 8.7: The data produced by the carbon dioxide sensors shows a Gaussian distribution.
Figure 8.8: Data collected from the methane sensors has been stacked in this graph to show the coincidence
of peaks in the concentration. Gaps represent communication errors with a single sensor subnet.
125
we can only take note of the shape of the normal distribution. The peak is Gaussian with
tailing to higher gas concentrations. This suggests that while most of the reported values
from the sensors fall within a small window, there are several events which cause this skew.
The kurtosis metric shows that the normal distribution is weighted more heavily toward the
mean.
Figure 8.9: The traces from the Farnsworth, TX site have been averaged in this graph to show a single line
which depicts the average concentration of methane at the site.
8.3.3 Peak Events
By applying some statistical analysis to the average of all sensors on the network
for both carbon dioxide and methane, we can determine when peak concentrations were
detected. These ‘Events’ were selected by determining if a value was greater than two
standard deviations above the average. This selection method removes the natural variation
in the diel cycle from generating a false positive. If we plot these events and their respective
concentrations against time, we can look for patterns, this plot is shown in Figure 8.11.
Looking at this plot, there are a few things which stand out as notable. First, there are broad
events in the methane plot. These suggest that the methane concentration was raised to
a high level and sustained for some time. Due to the multiple gas wells in the vicinity, it
is possible that these events suggest sustained releases from the wells. However, it is also
possible that this methane is due to anthropogenic gas production. The measurement period
coincided with two harvest periods for the local farmers in the region. It is possible that
126
Figure 8.10: The data produced by the methane sensors shows a Gaussian distribution.
the increased methane levels are due to the operation of poorly filtered engines used in farm
equipment. There are also a few events which line up directly between both gases. It is
possible that these events are due to releases from the injection wells. We are currently
communicating with the injection well company to determine if they have reported leaks
or releases during the time periods of both the gas-matching events and the long timescale
events.
8.4 Conclusions
The deployment of the sensor array networks near an injection well site have shown
several successes. The network has proved resilient to long distance data transfer and antenna
related downtime. The sensors have also reported two types of interesting events. The first is
127
2e-06
4e-06
6e-06
8e-06
1e-05
1.2e-05
1.4e-05
1.6e-05
09/01/16 10/01/16 11/01/16 12/01/16 01/01/17
Methane Conductivity (S)
0
200
400
600
800
1000
1200
1400
1600
Carbon Dioxide Conc. (ppm)
Figure 8.11: This plot shows the recorded events at the Farnsworth, TX site. Events were selected as
points which were 2 standard deviations higher than the average at that time of day.
a long term increase in the baseline methane concentration. The second are concentration
spikes that are detected by both carbon dioxide and methane sensors. It is possible that these
events are due to the release of gas from nearby wells, and the correlation with reported release
times is currently sought. This test demonstrates that the sensor network was successful in
monitoring the type of site which is vulnerable to simultaneous carbon dioxide and methane
bursts.
128
Chapter 9
Conclusion
9.1 Introduction
The goal of this project was to establish a wide-area sensing device using commercially
available sensors and equipment for application to carbon dioxide injection wells. The device
in question was a distributed system of networked units with appropriate sensors.
9.2 Summary of Work and Findings
Several commercially available sensors were compared for the determination of carbon
dioxide and methane concentrations. Sensors were selected based on low cost, low power
consumption, appreciable sensitivity near global concentration norms, and ready availability.
The sensitivity of the sensors was determined by gas exposure tests compared against a high
quality gas infrared instrument, the ZRE from California Analytical Instruments. In addition
to the sensitivity measurements performed for sensor selection purposes, the baseline noise,
limit of detection, and precision was determined for the available sensors.No single sensor for
either gas was able to satisfy the previously determined requirements for the selection of a
sensor in its entirety. Therefore, compromises were made based on the available technology.
Most units were equipped with the K-30 carbon dioxide sensor from CO2Meter and the MQ-4
chemiresistive methane sensor from Hanwei electronics. Certain sensors in the array were
equipped with a methane Gascard sensor from Edinburgh.
The particulars of certain elements of the sensor array were researched and documented
129
to facilitate design. With a solar power enclosure chosen, the characteristics of the batteries
were monitored at several different currents to check the cutoff voltage, observed capacity,
and cycling damage. Findings from these tests dictated certain measures be taken to ensure
battery longevity by preventing the charging circuit from using the manufacturer determined
cutoff point. The network was designed from available technology and theoretical work on
optimal networking strategies by other researchers. With this information, it was determined
that the most efficient model of moving the data from a field site to a home server was a
hybrid mesh network of short range wireless radios coupled to a cellular modem. Data were
moved through a ring buffer in the device code and synchronized with other units to add
redundancy to the network.
With the sensors selected and with the battery and network characterized, the devices
could be designed. Circuits were laid out which allowed for low power operation. The
sensors were implemented using a daughter board which could interface with the surrounding
environment in both passive sampling and flow regimes. For passive sampling, a 3D printed
interfacing part was designed which would allow for the sensitive electrical components to be
protected from unwanted environmental damage. Prototypes were built with these designs,
and veracity was confirmed. The devices were manufactured at scale and prepared for field
deployment.
Prototyping tests at the field site north of OSU campus provided for initial results
and baseline confirmation. Long term data were collected with the units to compare with
future work. The sensors were confirmed to be working by comparison with known weather
data from verified sources, and reasonable measurements of gas concentrations were obtained.
Sensors were deployed at two sites and collected data were analyzed from both as a
comparison. The first site was an unmanned aerial vehicle airport east of the OSU campus.
Data from this site were analyzed and treated as a control for future experiments. Peak
concentration events, determined by reported gas levels two standard deviations from the
mean, were observed. An experimental field site was established in around a carbon dioxide
injection well near Farnsworth, TX. This field site, when treated with the same statistical
conditions, was shown to have broader timescale peak events as well as events in which high
concentrations of both carbon dioxide and methane were observed coincident to each other.
Comparing the peak events at the Farnsworth, TX site to the UAS site, we can see that there
130
are distinct differences in the event patterns. This suggests that we may be detecting the gas
microseepage from the well head, the detection of which was the original goal of the project.
9.3 Beyond the Project Goals
With project goals met, it is possible to say that the implementation of a wide-area
distributed networked sensor array is a reasonable method of determining gas concentration
from areas below well heads. Further studies must be conducted to discover any possible
correlation of detected microseepage at the Farnsworth, TX site to geological changes or
geochemical processes occurring during the detection period. The current results suggest
that this array is useful as an early warning type monitoring system. Additional work may
determine that the capabilities of this array in conjunction with other technologies are suitable
for a more comprehensive monitoring strategy for these types of sites.
Distributed sensors have a very wide range of use for any application requiring detection
of properties over a large area or at multiple sites. Specifically, the technology developed
for this project has direct application to other research as well. Carbon dioxide is known to
leach from carbonaceous soils as a result of global warming in both temperate and permafrost
regions [
109
,
110
]. Additionally, tundra ice is known to release methane during thaw [
111
].
The sensor arrays could be deployed for measurement at these sites for long term monitoring.
The under-reporting of methane emissions by gathering and processing facilities [
112
] may be
partially remedied by deploying sensors at locations poorly monitored by the current standard,
such as wellheads, drilling, and exploration sites. Landfills produce a large amount of methane
leaking from the packed soil, currently measured by single FTIR sampling devices [
113
]. An
always-on array of sensors could monitor the output from the soil in real time to detect and
analyze fluctuations in this output. With small changes in hardware, the sensor array could
be adapted to volcanology applications. During eruption events, efforts are undertaken to
gather large quantities of data about the gaseous production of the volcano [
114
,
60
]. During
inactive periods however, only a few sensors are maintained to reduce cost and time spent
on measurement. A sensor array could monitor a large area of land for development of new
crevices and fissures producing gas without the need for an increase in labor.
The observations from this project are simple, yet they open doors to new possibilities
for environmental monitoring devices. The completion of this project provides a fundamental
131
framework for the design of future devices. In addition to the obvious continuation of the
project which promises more data and a continued stream of interesting observations, the
core design of the networked sensor array gives rise to further possible work in areas outside
of this project’s scope.
132
Bibliography
[1]
Svante Arrhenius. XXXI. On the influence of carbonic acid in the air upon the
temperature of the ground. Philosophical Magazine Series 5, 41(251):237–276, April
1896.
[2]
T. A. Boden, G. Marland, and R. J. Andres. Global, Regional, and National Fossil-Fuel
CO2 Emissions. Technical report, Carbon Dioxide Information Analysis Center, Oak
Ridge National Laboratory, U.S. Department of Energy, Oak Ridge, Tenn., U.S.A.,
2011.
[3]
Stefan Bachu. Sequestration of CO2 in geological media: criteria and approach for
site selection in response to climate change. Energy Conversion and Management,
41(9):953–970, June 2000.
[4]
Curt M. White, Brian R. Strazisar, Evan J. Granite, James S. Hoffman, and Henry W.
Pennline. Separation and Capture of CO2 from Large Stationary Sources and Seques-
tration in Geological FormationsCoalbeds and Deep Saline Aquifers. Journal of the Air
& Waste Management Association, 53(6):645–715, June 2003.
[5]
Hsien H. Khoo and Reginald B. H. Tan. Life Cycle Investigation of CO2 Recovery and
Sequestration. Environmental Science & Technology, 40(12):4016–4024, June 2006.
[6]
K. D. Romanak, P. C. Bennett, Changbing Yang, and Susan D. Hovorka. Process-based
approach to CO2 leakage detection by vadose zone gas monitoring at geologic CO2
storage sites. Geophysical Research Letters, 39(15):n/a–n/a, August 2012.
[7]
R.A. Chadwick, D. Noy, R. Arts, and O. Eiken. Latest time-lapse seismic data
from Sleipner yield new insights into CO2 plume development. Greenhouse Gas
Control Technologies 9Proceedings of the 9th International Conference on Greenhouse
133
Gas Control Technologies (GHGT-9), 1620 November 2008, Washington DC, USA,
1(1):2103–2110, February 2009.
[8]
Curt M. White, Duane H. Smith, Kenneth L. Jones, Angela L. Goodman, Sinisha A.
Jikich, Robert B. LaCount, Stephen B. DuBose, Ekrem Ozdemir, Badie I. Morsi, and
Karl T. Schroeder. Sequestration of Carbon Dioxide in Coal with Enhanced Coalbed
Methane RecoveryA Review. Energy & Fuels, 19(3):659–724, May 2005.
[9]
Peter Cook, Rick Causebrook, John Gale, Karsten Michel, and Max Watson. What
Have We Learned from Small-scale Injection Projects? 12th International Conference
on Greenhouse Gas Control Technologies, GHGT-12, 63:6129–6140, January 2014.
[10]
Vello A. Kuuskraa, Tyler Van Leeuwen, and Matt Wallace. Improving Domestic Energy
Security and Lowering CO2 Emissions with ”Next Generation” CO2-Enhanced Oil
Recovery (CO2-EOR). Technical Report DOE/NETL-2011/1504, National Energy
Technology Laboratory, June 2011.
[11]
Vello A. Kuuskraa, Michael L. Godec, and Phil Dipietro. CO2 Utilization from Next
Generation CO2 Enhanced Oil Recovery Technology. GHGT-11, 37:6854–6866, 2013.
[12]
Eduardo Jose Manrique, Viviana Eugenia Muci, and Mariano E. Gurfinkel. EOR Field
Experiences in Carbonate Reservoirs in the United States. In SPE-100063-MS, SPE,
January 2006. Society of Petroleum Engineers.
[13]
Yousif K. Kharaka, James J. Thordsen, Susan D. Hovorka, H. Seay Nance, David R.
Cole, Tommy J. Phelps, and Kevin G. Knauss. Potential environmental issues of CO2
storage in deep saline aquifers: Geochemical results from the Frio-I Brine Pilot test,
Texas, USA. IAGC Celebrates 40 Years Selected Papers from the 40th Anniversary
Celebration of the International Association of GeoChemistry, Cologne, Germany,
August 2007 and the Special Session at the 2007 Goldschmidt Conference in Memory
of A. A. Levinson, 24(6):1106–1112, June 2009.
[14]
George W. Kling, Michael A Clark, Glen N. Wagner, Harry R. Compton, Alan M.
Humphrey, Joseph D. Devine, William C. Evans, John P. Lockwood, Michelle L. Tuttle,
134
and Edward J. Koenigsberg. The 1986 Lake Nyos Gas Disaster in Cameroon, West
Africa. Science, 236(4798):169–175, April 1987.
[15]
H. Sigurdsson, J.D. Devine, F.M. Tchua, F.M. Presser, M.K.W. Pringle, and W.C.
Evans. Origin of the lethal gas burst from Lake Monoun, Cameroun. Journal of
Volcanology and Geothermal Research, 31(12):1–16, March 1987.
[16]
Ronald W. Klusman. Baseline studies of surface gas exchange and soil-gas compostion in
preparation for CO2 sequestration research: Teapot Dome, Wyoming. AAPG Bulletin,
89(8):981–1003, August 2005.
[17]
Ronald W Klusman. Rate measurements and detection of gas microseepage to the
atmosphere from an enhanced oil recovery/sequestration project, Rangely, Colorado,
USA. Applied Geochemistry, 18(12):1825–1838, December 2003.
[18]
Susan Carrroll, Yue Hao, and Roger Aines. Geochemical detection of carbon dioxide in
dilute aquifers. Geochemical Transactions, 10(4), March 2009.
[19]
Vincent Vandeweijer, Bert van der Meer, Cor Hofstee, Frans Mulders, Daan DHoore,
and Hilbrand Graven. Monitoring the CO2 injection site: K12-B. 10th International
Conference on Greenhouse Gas Control Technologies, 4:5471–5478, January 2011.
[20]
C. M. Oldenburg, K. Pruess, and S. M. Benson. Process Modeling of CO2 Injection
into Natural Gas Reservoirs for Carbon Sequestration and Enhanced Gas Recovery.
Energy & Fuels, 15(2):293–298, March 2001.
[21]
X. G. Zhang, P. G. Ranjith, M. S. A. Perera, A. S. Ranathunga, and A. Haque. Gas
Transportation and Enhanced Coalbed Methane Recovery Processes in Deep Coal
Seams: A Review. Energy & Fuels, 30(11):8832–8849, November 2016.
[22]
A. K. Singh, G. Baumann, J. Henninges, U.-J. Grke, and O. Kolditz. Numerical
analysis of thermal effects during carbon dioxide injection with enhanced gas recovery:
a theoretical case study for the Altmark gas field. Environmental Earth Sciences,
67(2):497–509, 2012.
135
[23]
Abdolvahab Honari, Branko Bijeljic, Michael L. Johns, and Eric F. May. Enhanced gas
recovery with CO2 sequestration: The effect of medium heterogeneity on the dispersion
of supercritical CO2CH4. International J ournal of Greenhouse Gas Control, 39:39–50,
August 2015.
[24]
Milan J. Patel, Eric F. May, and Michael L. Johns. High-fidelity reservoir simulations of
enhanced gas recovery with supercritical CO2. Energy, 111:548–559, September 2016.
[25]
NETL. Best Practices for Monitoring, Verification, and Accounting of CO2 Stored
in Deep Geologic Formations - 2012 Update. Best Practices Report DOE/NETL-
311/081508, U. S. Department of Energy, National Energy Technology Laboratory,
2012.
[26]
L. Bateson, M. Vellico, S.E. Beaubien, J.M. Pearce, A. Annunziatellis, G. Ciotoli,
F. Coren, S. Lombardi, and S. Marsh. The application of remote-sensing techniques
to monitor CO2-storage sites for surface leakage: Method development and testing
at Latera (Italy) where naturally produced CO2 is leaking to the atmosphere. EGU
General Assembly 2007: Advances in CO2 Storage in Geological SystemsEGU (2007),
2(3):388–400, July 2008.
[27]
E.J. Male, W.L. Pickles, E.A. Silver, G.D. Hoffmann, J. Lewicki, M. Apple, K. Repasky,
and E.A. Burton. Using hyperspectral plant signatures for CO2 leak detection during the
2008 ZERT CO2 sequestration field experiment in Bozeman, Montana. Environmental
Earth Sciences, November 2009.
[28]
Esther Salam, Cristina Barrado, and Enric Pastor. UAV Flight Experiments Applied to
the Remote Sensing of Vegetated Areas. Remote Sensing, 6(11):11051–11081, November
2014.
[29]
D. R. Thompson, A. K. Thorpe, C. Frankenberg, R. O. Green, R. Duren, L. Guanter,
A. Hollstein, E. Middleton, L. Ong, and S. Ungar. Space-based remote imaging
spectroscopy of the Aliso Canyon CH4 superemitter. Geophysical Research Letters,
43(12):6571–6578, 2016.
136
[30]
Jamie L. Barr, Seth D. Humphries, Amin R. Nehrir, Kevin S. Repasky, Laura M.
Dobeck, John L. Carlsten, and Lee H. Spangler. Laser-based carbon dioxide moni-
toring instrument testing during a 30-day controlled underground carbon release field
experiment. International Journal of Greenhouse Gas Control, 5(1):138–145, January
2011.
[31]
Seth D. Humphries, Amin R. Nehrir, Charlie J. Keith, Kevin S. Repasky, Laura M.
Dobeck, John L. Carlsten, and Lee H. Spangler. Testing carbon sequestration site
monitor instruments using a controlled carbon dioxide release facility. Applied Optics,
47(4):548–555, February 2008.
[32]
Jan M. Nordbotten, Dmitri Kavetski, Michael A. Celia, and Stefan Bachu. Model
for CO2 Leakage Including Multiple Geological Layers and Multiple Leaky Wells.
Environmental Science & Technology, 43(3):743–749, February 2009.
[33]
Jan Martin Nordbotten, Michael A. Celia, Stefan Bachu, and Helge K. Dahle. Semian-
alytical Solution for CO2 Leakage through an Abandoned Well. Environmental Science
& Technology, 39(2):602–611, January 2005.
[34]
Ronald W. Klusman. Comparison of surface and near-surface geochemical methods for
detection of gas microseepage from carbon dioxide sequestration. International Journal
of Greenhouse Gas Control, 5(6):1369–1392, November 2011.
[35]
Arthur W. Wells, J. Rodney Diehl, Brian R. Strazisar, Thomas H. Wilson, and Dennis C.
Stanko. Atmospheric and soil-gas monitoring for surface leakage at the San Juan Basin
CO2 pilot test site at Pump Canyon New Mexico, using perfluorocarbon tracers, CO2
soil-gas flux and soil-gas hydrocarbons. International Journal of Greenhouse Gas
Control, 14:227–238, May 2013.
[36]
Arthur W. Wells, J. Rodney Diehl, Grant Bromhal, Brian R. Strazisar, Thomas H.
Wilson, and Curt M. White. The use of tracers to assess leakage from the sequestration
of CO2 in a depleted oil reservoir, New Mexico, USA. Applied Geochemistry, 22(5):996–
1016, May 2007.
137
[37]
F.F. Craig, III. Field Use of Halogen Compounds To Trace Injected CO2. In SPE-
14309-MS, SPE, January 1985. Society of Petroleum Engineers.
[38]
G. Myhre, D. Shindell, F.-M. Bron, W. Collins, J. Fuglestvedt, J. Huang, D. Koch, J.-F.
Lemarque, D. Lee, B. Mendoza, T. Nakajima, A. Robock, G. Stephens, T. Takemura,
and H. Zhang. 2013: Anthropogenic and Natural Radiative Forcing Supplementary
Material. In: Climate Change 2013: The Physical Science Basis. Contribution of
Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on
Climate Change. Technical report, 2013.
[39]
J.A. Apps. A Review of Hazardous Chemical Species Associated with CO2 Capturefrom
Coal-Fired Power Plants and Their Potential Fate in CO2 GeologicStorage. Technical
report, United States, February 2006. DOI: 10.2172/888971.
[40]
Peter A. Raymond, Nina F. Caraco, and Jonathan J. Cole. Carbon dioxide concentration
and atmospheric flux in the Hudson River. Estuaries, 20(2):381–390, 1997.
[41]
Michael L. Goulden, Scott D. Miller, Humberto R. da Rocha, Mary C. Menton,
Helber C. de Freitas, Adelaine Michela e Silva Figueira, and Cleilim Albert Dias
de Sousa. DIEL AND SEASONAL PATTERNS OF TROPICAL FOREST CO2
EXCHANGE. Ecological Applications, 14(sp4):42–54, August 2004.
[42]
J. W. Raich, R. D. Bowden, and P. A. Steudler. Comparison of Two Static Chamber
Techniques for Determining Carbon Dioxide Efflux from Forest Soils. Soil Science
Society of America Journal, 54(6):1754–1757, 1990.
[43]
SEIKO OSOZAWA and SHUICHI HASEGAWA. DIEL AND SEASONAL CHANGES
IN CARBON DIOXIDE CONCENTRATION AND FLUX IN AN ANDISOL. Soil
Science, 160(2), 1995.
[44]
Ji-Qin Ni, Albert J. Heber, Claude A. Diehl, and Teng T. Lim. SEStructures and
Environment. Journal of Agricultural Engineering Research, 77(1):53–66, September
2000.
[45]
S.C. MABERLY. Diel, episodic and seasonal changes in pH and concentrations of
inorganic carbon in a productive lake. Freshwater Biology, 35(3):579–598, 1996.
138
[46]
Weixin Ding, Zucong Cai, and Haruo Tsuruta. Diel variation in methane emissions
from the stands of Carex lasiocarpa and Deyeuxia angustifolia in a cool temperate
freshwater marsh. Atmospheric Environment, 38(2):181–188, January 2004.
[47]
Tiina Kki, Anne Ojala, and Paula Kankaala. Diel variation in methane emissions from
stands of Phragmites australis (Cav.) Trin. ex Steud. and Typha latifolia L. in a boreal
lake. Aquatic Botany, 71(4):259–271, December 2001.
[48]
Yaohong Zhang and Weixin Ding. Diel methane emissions in stands of Spartina
alterniflora and Suaeda salsa from a coastal salt marsh. Aquatic Botany, 95(4):262–267,
November 2011.
[49]
B. Wang, H.U. Neue, and H.P. Samonte. The effect of controlled soil temperature on
diel CH4 emission variation. Chemosphere, 35(9):2083–2092, November 1997.
[50]
B. Wang, H.U. Neue, and H.P. Samonte. Factors controlling diel patterns of methane
emission via rice. Nutrient Cycling in Agroecosystems, 53(3):229–235, 1999.
[51]
Frans-Faco W.A. Van Der Nat, Jack J. Middelburg*, Danille Van Meteren, and Annette
Wielemakers. Diel methane emission patterns from Scirpus lacustris and Phragmites
australis. Biogeochemistry, 41(1):1–22, 1998.
[52]
Antonio-Javier Garcia-Sanchez, Felipe Garcia-Sanchez, and Joan Garcia-Haro. Wireless
sensor network deployment for integrating video-surveillance and data-monitoring in
precision agriculture over distributed crops. Computers and Electronics in Agriculture,
75(2):288–303, February 2011.
[53]
Jude Allred, Ahmad Bilal Hasan, Saroch Panichsakul, William Pisano, Peter Gray, Jyh
Huang, Richard Han, Dale Lawrence, and Kamran Mohseni. SensorFlock: An Airborne
Wireless Sensor Network of Micro-air Vehicles. In Proceedings of the 5th International
Conference on Embedded Networked Sensor Systems, SenSys ’07, pages 117–129, New
York, NY, USA, 2007. ACM.
[54]
Anna Korre, Claire E. Imrie, Franz May, Stan E. Beaubien, Vincent Vandermeijer, Sergio
Persoglia, Lars Golmen, Hubert Fabriol, and Tim Dixon. Quantification techniques for
139
potential CO2 leakage from geological storage sites. 10th International Conference on
Greenhouse Gas Control Technologies, 4:3413–3420, 2011.
[55]
Zheng Yang, Nan Li, Burcin Becerik-Gerber, and Michael Orosz. A systematic approach
to occupancy modeling in ambient sensor-rich buildings. SIMULATION, 90(8):960–977,
August 2014. bibtex: yang systematic 2014.
[56]
Wan-Young Chung and Seung-Chul Lee. A selective AQS system with artificial
neural network in automobile. Proceedings of the Eleventh International Meeting on
Chemical Sensors IMCS-11IMCS 2006IMCS 11, 130(1):258–263, March 2008. bibtex:
chung selective 2008.
[57]
Wangyun Won and Kwang Soon Lee. Nonlinear observer with adaptive grid allocation
for a fixed-bed adsorption process. Computers & Chemical Engineering, 46:69–77,
November 2012.
[58]
Pan Yi, L. Xiao, and Y. Zhang. Remote real-time monitoring system for oil and gas well
based on wireless sensor networks. In Mechanic Automation and Control Engineering
(MACE), 2010 International Conference on, pages 2427–2429, June 2010. bibtex:
yi remote 2010.
[59]
Andrey Somov, Alexander Baranov, Denis Spirjakin, Andrey Spirjakin, Vladimir
Sleptsov, and Roberto Passerone. Deployment and evaluation of a wireless sensor net-
work for methane leak detection. Selected Papers from the 26th European Conference on
Solid-State Transducers Krakw, Poland, 9-12 September 2012, 202:217–225, November
2013. bibtex: somov deployment 2013.
[60]
T.D. Pering, G. Tamburello, A.J.S. McGonigle, A. Aiuppa, A. Cannata, G. Giudice,
and D. Patan. High time resolution fluctuations in volcanic carbon dioxide degassing
from Mount Etna. Journal of Volcanology and Geothermal Research, 270:115–121,
January 2014.
[61]
R.R. Black, C.P. (Mick) Meyer, A. Yates, L. Van Zweiten, and J.F. Mueller. Formation
of artefacts while sampling emissions of PCDD/PCDF from open burning of biomass.
Chemosphere, 88(3):352–357, July 2012.
140
[62]
Hui Guohua, Wang Lvye, Mo Yanhong, and Zhang Lingxia. Study of grass carp
(Ctenopharyngodon idellus) quality predictive model based on electronic nose. Sensors
and Actuators B: Chemical, 166167:301–308, May 2012.
[63]
S. Karunanithi, N. M. Din, H. Hakimie, C. K. Hua, R. C. Omar, and T. C. Yee.
Performance of labscale solar powered wireless landfill monitoring system. In Energy
and Environment, 2009. ICEE 2009. 3rd International Conference on, pages 443–448,
December 2009. bibtex: karunanithi performance 2009.
[64]
Derek G. Shendell, Jennifer H. Therkorn, Naomichi Yamamoto, Qingyu Meng, Sarah W.
Kelly, and Christine A. Foster. Outdoor near-roadway, community and residential
pollen, carbon dioxide and particulate matter measurements in the urban core of an
agricultural region in central CA. Atmospheric Environment, 50:103–111, April 2012.
[65]
Shih-Wen Chiu and Kea-Tiong Tang. Towards a Chemiresistive Sensor-Integrated
Electronic Nose: A Review. Sensors, 13(10):14214–14247, October 2013.
[66]
Clifford K. Ho, Michael T. Itamura, Michael Kelley, and Robert C. Hughes. Review of
Chemical Sensors for Real-Time In-Situ Sensing. Technical Report SAND2001-0643,
Sandia National Laboratories, 2001.
[67]
T Blasing. Recent Greenhouse Gas Concentrations. Technical report, U. S. Department
of Energy, CDIAC, 2016.
[68]
Ed Dlugokencky and Pieter Tans. Trends in Atmospheric Carbon Dioxide. Technical
report, NOAA/ESRL, 2016.
[69]
A. J. Turner, D. J. Jacob, J. Benmergui, S. C. Wofsy, J. D. Maasakkers, A. Butz,
O. Hasekamp, and S. C. Biraud. A large increase in U.S. methane emissions over the
past decade inferred from satellite data and surface observations. Geophysical Research
Letters, 43(5):2218–2224, March 2016.
[70]
I. Bamberger, J. Stieger, N. Buchmann, and W. Eugster. Spatial variability of methane:
Attributing atmospheric concentrations to emissions. Environmental Pollution , 190:65–
74, July 2014.
141
[71]
Ed Dlugokencky. Trends in Atmospheric Methane. Technical report, NOAA/ESRL,
2016.
[72]
Giovanni Neri. First Fifty Years of Chemoresistive Gas Sensors. Chemosensors, 3(1):1–
20, January 2015.
[73]
R. Frodl and T. Tille. A High-Precision NDIR Gas Sensor for Automotive Applications.
IEEE Sensors Journal, 6(6):1697–1705, December 2006.
[74]
Zipeng Zhu, Yuhui Xu, and Binqing Jiang. A One ppm NDIR Methane Gas Sensor with
Single Frequency Filter Denoising Algorithm. Sensors (Basel, Switzerland), 12(9):12729–
12740, 2012.
[75]
Zoltn Bacsik, Jnos Mink, and Gbor Keresztury. FTIR Spectroscopy of the Atmosphere.
I. Principles and Methods. Applied Spectroscopy Reviews, 39(3):295–363, December
2004.
[76]
Coblentz Society, Inc. Evaluated Infrared Reference Spectra. In P. J. Lindstrom and
W. G. Mallard, editors, NIST Chemistry WebBook, number 69 in NIST Standard
Reference Database. National Institute of Standards and Technology, Gaithersburg
MD, 20899.
[77]
Keith J. Albert, Nathan S. Lewis, Caroline L. Schauer, Gregory A. Sotzing, Shannon E.
Stitzel, Thomas P. Vaid, and David R. Walt. Cross-Reactive Chemical Sensor Arrays.
Chemical Reviews, 100(7):2595–2626, July 2000.
[78]
Chengxiang Wang, Longwei Yin, Luyuan Zhang, Dong Xiang, and Rui Gao. Metal
oxide gas sensors: sensitivity and influencing factors. Sensors, 10(3):2088–2106, 2010.
[79]
Maria Prudenziati and Bruno Morten. Thick-film sensors: an overview. Sensors and
Actuators, 10(1):65–82, September 1986.
[80]
Corrado Di Natale, Fabrizio Davide, Guido Faglia, and Paolo Nelli. Study of the effect
of the sensor operating temperature on SnO2-based sensor-array performance. The
workshop on new developments in semiconducting gas sensors, 23(2):187–191, February
1995.
142
[81]
Ling Wang and R.V. Kumar. Thick film CO2 sensors based on Nasicon solid electrolyte.
Solid State Ionics, 158(34):309–315, March 2003.
[82]
W. Eugster and G. W. Kling. Performance of a low-cost methane sensor for ambient con-
centration measurements in preliminary studies. Atmospheric Measurement Techniques,
5(Copyright (C) 2015 American Chemical Society (ACS). All Rights Reserved.):1925–
1934, 2012.
[83]
Lawrence H. Keith, Warren Crummett, John Deegan, Robert A. Libby, John K. Taylor,
and George Wentler. Principles of environmental analysis. Analytical Chemistry,
55(14):2210–2218, 1983.
[84]
Mocak J., Bond A. M., Mitchell S., and Scollary G. A statistical overview of standard
(IUPAC and ACS) and new procedures for determining the limits of detection and
quantification: Application to voltammetric and stripping techniques (Technical Report).
Pure and Applied Chemistry, 69(2):297, 2009.
[85]
T. Williams and C. Kelley. gnuplot 5.0: An Interactive Plotting Program, January
2016.
[86] Tycon Power Systems. RemotePro Data Sheet, November 2014.
[87]
Ramin Hekmat. Ad-hoc Networks: Funadamental Properties and Network Topologies.
Springer Netherlands, 2006.
[88]
Victor O. K. Li, Y. F. Lam, T. C. Hou, and Joesph H. Yuen. Topology Design and
Performance Analysis of an Integrated Communication Network. JPL Publication JPL
Pub. 85-97, California Institute of Technology, Jet Propulsion Laboratory, September
1985.
[89]
FCC ID: MCQ-XB900hp. FCC License MCQ-XB900HP, Federal Communications
Commission, 2012.
[90]
IEEE Standard for Local and metropolitan area networks Part 15.4: Low-Rate Wireless
Personal Area Networks (LR-WPANs). Technical Report IEEE 802.15.7, The Institute
of Electrical and Electronics Engineers, Inc., 2011.
143
[91] Wireless Mesh Networking ZigBee vs. DigiMesh. Technical report, 2008.
[92]
Robert Faludi. Building wireless sensor networks: with ZigBee, XBee, arduino, and
processing. ” O’Reilly Media, Inc.”, 2010.
[93]
Qualcomm Incorporated. EV-DO Rev. A and B: Wireless Broadband for the Masses.
Technical report, December 2007.
[94]
KSWO. Weather History for KSWO from July 1, 2014 to July 31, 2015 [Dataset], 2015.
[95]
M Kintel and C Clifford Wolf. OpenSCAD, The Programmers Solid 3d CAD Modeller.
2011.
[96] KSWO. Weather History for KSWO from September 11, 2015 [Dataset], 2015.
[97] KSWO. Weather History for KSWO from January 1, 2016 [Dataset], 2016.
[98]
M. G. Tarnawski, T. G. A. Green, B. Buedel, A. Meyer, H. Zellner, and O. L. Lange.
Diel changes of atmospheric CO2 concentration within, and above, cryptogam stands
in a New Zealand temperate rainforest. New Zealand Journal of Botany, 32(3):329–336,
July 1994.
[99]
Seok-In Yun, Woo-Jung Choi, Jae-Eul Choi, and Han-Yong Kim. High-Time Res-
olution Analysis of Diel Variation in Methane Emission from Flooded Rice Fields.
Communications in Soil Science and Plant Analysis, 44(10):1620–1628, May 2013.
[100]
D. Y. F. Lai, N. T. Roulet, E. R. Humphreys, T. R. Moore, and M. Dalva. The effect
of atmospheric turbulence and chamber deployment period on autochamber CO
2
and
CH
4
flux measurements in an ombrotrophic peatland. Biogeosciences, 9(8):3305–3322,
2012.
[101]
S. M. McGinn, K. A. Beauchemin, T. Coates, and E. J. McGeough. Cattle Methane
Emission and Pasture Carbon Dioxide Balance of a Grazed Grassland. Journal of
Environmental Quality, 43(3):820–828, 2014.
[102]
Mahlon M. Ball, Mitchell E. Henry, and Sherwood E. Frezon. Petroleum Geology of
the Anadarko Basin Region, Province (115), Kansas, Oklahoma, and Texas. USGS
Survey 88-450W, USGS, 1991.
144
[103]
M.D. White, B.J. McPherson, R.B. Grigg, W. Ampomah, and M.S. Appold. Numerical
Simulation of Carbon Dioxide Injection in the Western Section of the Farnsworth Unit.
12th International Conference on Greenhouse Gas Control Technologies, GHGT-12,
63:7891–7912, 2014.
[104]
NETL. New Approach for Long Term Monitoring of Potential CO2 Leaks from Geologic
Sequestration. Project Fact Sheet FWPAACH139, U. S. Department of Energy, National
Energy Technology Laboratory, 2010.
[105]
NETL. Site Characterization for CO2 Storage from Coal-fired Power Facilities in the
Black Warrior Basin of Alabama. Project Fact Sheet FE0001910, U. S. Department of
Energy, National Energy Technology Laboratory, 2012.
[106] N. Materer, A. Apblett, and D. Scott. Chlorine dioxide sensor, August 2010.
[107]
N. Materer, P. Field, N. Ley, A. Soufiani, D. Scott, T. Ley, and A. Apblett. Passive
Wireless Detection of Corrosive Salts in Concrete Using Wire-Based Triggers. Journal
of Materials in Civil Engineering, 26(5):918–922, June 2013.
[108]
M. Pour-Ghaz, T. Barrett, T. Ley, N. Materer, A. Apblett, and J. Weiss. Wireless Crack
Detection in Concrete Elements Using Conductive Surface Sensors and Radio Frequency
Identification Technology. Journal of Materials in Civil Engineering, 26(5):923–929,
July 2013.
[109]
D. S. Jenkinson, D. E. Adams, and A. Wild. Model estimates of CO2 emissions from
soil in response to global warming. Nature, 351(6324):304–306, May 1991.
[110]
Edward A. G. Schuur, Jason G. Vogel, Kathryn G. Crummer, Hanna Lee, James O.
Sickman, and T. E. Osterkamp. The effect of permafrost thaw on old carbon release
and net carbon exchange from tundra. Nature, 459(7246):556–559, May 2009.
[111]
Stephen C. Whalen and William S. Reeburgh. A methane flux time series for tundra
environments. Global Biogeochemical Cycles, 2(4):399–409, December 1988.
[112]
Anthony J. Marchese, Timothy L. Vaughn, Daniel J. Zimmerle, David M. Martinez,
Laurie L. Williams, Allen L. Robinson, Austin L. Mitchell, R. Subramanian, Daniel S.
145
Tkacik, Joseph R. Roscioli, and Scott C. Herndon. Methane Emissions from United
States Natural Gas Gathering and Processing. Environmental Science & Technology,
49(17):10718–10727, September 2015.
[113]
Bo Galle, Jerker Samuelsson, Bo H. Svensson, and Gunnar Brjesson. Measurements
of Methane Emissions from Landfills Using a Time Correlation Tracer Method Based
on FTIR Absorption Spectroscopy. Environmental Science & Technology, 35(1):21–25,
January 2001.
[114]
Hayley J Duffell, Clive Oppenheimer, David M Pyle, Bo Galle, Andrew J.S McGonigle,
and Mike R Burton. Changes in gas composition prior to a minor explosive eruption
at Masaya volcano, Nicaragua. Journal of Volcanology and Geothermal Research,
126(34):327–339, August 2003.
146
Appendix A
Sensor Board - Primary Circuit
147
148
149
150
151
152
153
154
155
156
157
158
159
160
Appendix B
Sensor Board - Sensors Breakout Circuit
161
162
163
164
165
Appendix C
Sensor Board - Cellular Modem Breakout Circuit
166
Appendix D
OpenSCAD Code Used for 3D Printed Part
Listing D.1: OpenSCAD script which generates plastic housing 3D model
module Requ iredBit ()
{
$ f s = 0 . 0 1 ;
d i f f e r e n c e ( )
{
union ( )
{
// main box
t r a n s l a t e ( [ −. 4 , −.0878 , 0 ] )
cube ( [ 3 . 8 7 5 , 2 . 1 5 , . 2 0 8 ] , c e n t e r = tr ue ) ;
t r a n s l a t e ( [ 1 . 0 3 6 5 , 0 , . 3 5 1 ] )
c y l i n d e r ( h = . 4 94 , r = . 2 7 5 , c e n t e r = tr ue ) ;
t r a n s l a t e ([ −. 6 565 , 0 , . 3 5 1 ] )
c y l i n d e r ( h = . 4 94 , r = . 2 5 2 5 , c e n t e r = t r u e ) ;
// extrabottom
t r a n s l a t e ([ − . 50 0 , 0 , −.106])
cube ( [ 2 . 0 7 5 , 1 . 7 5 2 , . 0 5 4 ] , c e n t e r = tru e ) ;
167
}
// s t o v e p i p e cut throughs
t r a n s l a t e ( [ 1 . 0 3 6 5 , 0 , . 3 5 1 ] )
c y l i n d e r ( h = 1. 0 0 0 , r = . 1 8 5 , c e n t e r = tr ue ) ;
t r a n s l a t e ([ −. 6 565 , 0 , . 3 5 1 ] )
c y l i n d e r ( h = 1. 2 0 0 , r = . 1 8 5 , c e n t e r = tr ue ) ;
// s c r e wh ol e 1 a ( boxend , notch )
t r a n s l a t e ( [ 1 . 0 3 6 5 , . 6 5 0 , −.039])
c y l i n d e r ( h = . 1 30 , r = . 0 6 7 5 , c e n t e r = t r u e ) ;
t r a n s l a t e ( [ 1 . 0 3 6 5 , . 6 5 0 , . 0 6 5 ] )
c y l i n d e r ( h = . 0 78 , r = . 1 2 , c e n t e r = tr u e ) ;
// s c r e wh ol e 1 b ( boxend , top )
t r a n s l a t e ( [ 1 . 0 3 6 5 , −.670 , −. 039])
c y l i n d e r ( h = . 2 30 , r = . 0 6 7 5 , c e n t e r = t r u e ) ;
t r a n s l a t e ( [ 1 . 0 3 6 5 , −.670 , . 0 6 5 ] )
c y l i n d e r ( h = . 0 78 , r = . 1 2 , c e n t e r = tr u e ) ;
// s c r e wh ol e 2 a ( fr e ee nd , notch )
t r a n s l a t e ([ −2.1 761 , . 8 3 5 4 , −.039])
c y l i n d e r ( h = . 2 30 , r = . 0 6 7 5 , c e n t e r = t r u e ) ;
t r a n s l a t e ([ −2.1 761 , . 8 3 5 4 , . 0 6 5 ] )
c y l i n d e r ( h = . 0 78 , r = . 1 2 , c e n t e r = tr u e ) ;
// s c r e wh ol e 2 b ( fr eee nd , top )
t r a n s l a t e ([ −2.1 761 , −.9968 , −.039])
c y l i n d e r ( h = . 2 30 , r = . 0 6 7 5 , c e n t e r = t r u e ) ;
t r a n s l a t e ([ −2.1 761 , −.9968 , . 0 6 5 ] )
c y l i n d e r ( h = . 0 78 , r = . 1 2 , c e n t e r = tr u e ) ;
// antenna notch
t r a n s l a t e ([ −2.0 525 , 0 , 0 ] )
c y l i n d e r ( h = 1 , r = .3 1 2 5 , c e nt e r = tru e ) ;
t r a n s l a t e ([ −2.0 525 , 0 . 08 , 0 ] )
c y l i n d e r ( h = 1 , r = .3 1 2 5 , c e nt e r = tru e ) ;
168
t r a n s l a t e ([ − 2 .0 5 , 0 . 0 4 , 0 ] )
c y l i n d e r ( h = 1 , r = .3 1 2 5 , c e nt e r = tru e ) ;
// an t enna pret ty − c l e a n s up antenna ho le
t r a n s l a t e ([ −2 . 365 , 0 , 0 ] )
cube ( [ . 6 2 5 , . 6 2 5 , 1 ] , c e n t e r = tru e ) ;
t r a n s l a t e ([ −2 . 365 , 0 . 0 8 , 0 ] )
cube ( [ . 6 2 5 , . 6 2 5 , 1 ] , c e n t e r = tru e ) ;
// vent − adds notch f o r vent hol e on e n c l o s u r e
t r a n s l a t e ([ − 2. 3 , −0.4 , 0 ] )
c y l i n d e r ( h = 1 , r = . 2 5 , c e n t e r = tru e ) ;
t r a n s l a t e ([ − 2. 1 , −0.23 , 0 ] )
r o t a t e ( 5 1 . 5 )
cube ( [ . 5 , . 3 9 6 , 1 ] , c e n te r = tru e ) ;
// notch − to accommodate en c l o s u r e shape
t r a n s l a t e ([ −1.4 375 , . 8 , 0 ] )
c y l i n d e r ( h = 1 , r = .3 8 7 5 , c e nt e r = tru e ) ;
t r a n s l a t e ([ −1.4 375 , 1 . 2 , 0 ] )
cube ( [ . 7 7 5 , . 7 7 5 , 1 ] , c e n t e r = tru e ) ;
// b i g d i g − red uc e s m a t e r i a l used in part
t r a n s l a t e ([ − . 58 5 , −1.2 , 0 ] )
cube ( [ 2 , . 8 , 1 ] , c e n t e r = tr ue ) ;
t r a n s l a t e ([ − 1 .5 5 , −1.2 , 0 ] )
c y l i n d e r ( h = 1 , r = . 4 , c e n t e r = t r u e ) ;
t r a n s l a t e ( [ . 7 8 5 , −1.2 , 0 ] )
cube ( [ 2 , . 6 , 1 ] , c e n t e r = tr ue ) ;
// d e ep er d ig − r ed u c e s m a t e r i a l used in part
r o t a t e (4 2)
t r a n s l a t e ([ − .2 5 , −.8 , 0 ] )
s c a l e ( v =[1 , 2 , 1 ] )
c y l i n d e r ( h = 1 , r = . 3 , c e n t e r = t r u e ) ;
r o t a t e (13 8)
169
t r a n s l a t e ( [ −. 4 , . 5 9 , 0 ] )
s c a l e ( v =[1 , 1 . 7 , 1 ] )
c y l i n d e r ( h = 1 , r = . 3 , c e n t e r = t r u e ) ;
// lower po r t i o n c u t o f f / b e v e l − accommodates f i l l e t i n e n c l o s u r e
t r a n s l a t e ( [ 0 , 1 . 01 , 0 ] )
cube ( [ 3 . 3 , . 2 5 , 1 ] , c e n t e r = t r ue ) ;
// lower screw− s t r e n g t h e n s screw ho l e near c u t o f f / b e ve l
d i f f e r e n c e ( )
{
t r a n s l a t e ([ −2.1 761 , 1 , . 0 4 ] )
cube ( [ 1 , . 2 5 , . 1 3 ] , c e n t e r = tr u e ) ;
t r a n s l a t e ([ −2.1 761 , . 8 3 5 4 , 0 ] )
c y l i n d e r ( h = . 0 78 , r = . 1 2 , c e n t e r = tr u e ) ;
}
// lower c u t o f f / b ev el 2 − accommodates f i l l e t in e n c l o s u r e
t r a n s l a t e ([ −2.1 761 , 1 . 08 , −.065])
cube ( [ 1 , . 2 5 , . 0 8 1 ] , c e n t e r = t r ue ) ;
}
}
module OptionalBox ( )
{
t r a n s l a t e ( [ 1 . 0 3 6 5 , 0 , −.4165])
d i f f e r e n c e ( )
{
cube ( [ 1 . 0 0 0 , 1 . 0 0 0 , . 6 2 5 ] , c e n t e r = tru e ) ;
cube ( [ . 9 1 0 , . 9 1 0 , . 6 2 5 ] , c e n t e r = t r u e ) ;
}
}
// OptionalBox adds i s o l a t i n g box around p re ss u re and temperature s e n s o r s .
170
OptionalBox ( ) ;
RequiredBi t ( ) ;
171
Appendix E
Assembly Instructions For Sensor Nodes
172
Figure E.1: Instructions for sensor node assembly, slides 1-6.
173
Figure E.2: Instructions for sensor node assembly, slides 7-12.
174
Figure E.3: Instructions for sensor node assembly, slides 13-18.
175
Figure E.4: Instructions for sensor node assembly, slides 19-24.
176
Figure E.5: Instructions for sensor node assembly, slides 25-30.
177
Figure E.6: Instructions for sensor node assembly, slides 31-36.
178
Figure E.7: Instructions for sensor node assembly, slides 37-42.
179
Figure E.8: Instructions for sensor node assembly, slides 43-48.
180
Figure E.9: Instructions for sensor node assembly, slides 49-54.
181
Figure E.10: Instructions for sensor node assembly, slides 55-60.
182
VITA
Wesley T. Honeycutt
Candidate for the Degree of:
Doctor of Philosophy
Thesis: Development and Applications of Chemical Sensors for the Detection
of Atmospheric Carbon Dioxide and Methane
Major Field: Chemistry
Biographical:
Education:
Completed the requirements for Doctor of Philosophy in Chemistry at Oklahoma
State University, Stillwater Oklahoma in May, 2017.
Completed the requirements for Bachelor of Science in Chemistry at University of
Oklahoma, Norman Oklahoma in 2011.
Experience:
Employed by Oklahoma State Univeristy in the position of Research/Teaching Assis-
tant in Stillwater, Oklahoma from August 2012 to date.
Employed by ChevronPhillips Chemical Company in the position of Polymer Techni-
cian in Bartlesville, Oklahoma from July 2011 to March 2012.
Professional Memberships:
President of OSU Phi Lambda Upsilon as of August, 2016.
