1

ADVANCED SENSORS, WIRELESS NETWORKING, AND DATA ANALYSIS FOR LARGE-SCALE AQUIFER

CHARACTERIZATION AND MONITORING

Report on Project Funded by Stanford Wood’s and Energy and Environment Affiliates Program “Using Sensors and Big Data

to Solve Environmental Challenges”

PIs: Peter K. Kitanidis (Civil and Environmental Engineering), Thomas W. Kenny (Mechanical Engineering)

SUMMARY In this project, we advanced a technology for monitoring aquifer hydraulic dynamics, using modern microfabricated sensors, simplified data acquisition, wireless networking, and improved algorithms for tomography. We have developed a wireless data acquisition system with fiber-optic technology pressure transducers and have already deployed this system in filed work. In parallel, we have been making progress in adapting miniature and low-cost microfabricated pressure sensors to conditions of exposure and pressure anticipated in groundwater applications. We have also been working on the development of signal processing and inverse modeling techniques that allow us to utilize large quantities of data, both produced from sources such as periodic variations of head in wells and random ambient signals with a broad spectrum of fluctuations, to characterize parameters of hydrogeologic system, like conductivity and storage coefficients, as well as changes in the state of the system, like pressure. INTRODUCTION Freshwater is the critical resource for life on this planet, and groundwater represents the largest source of freshwater for human use. With populations increasing and developing economically, the demand for high quality freshwater is rising. A changing climate is decreasing the availability and reliability of freshwater sources around which societies have built institutions and infrastructure. In California dams, pipes and canals have been built to store and transfer water from precipitation in the Sierra Nevada Mountains to the population centers along the coast and to agriculture in the Central Valley. This system depends greatly on snowpack in the Sierras storing water and making it available when it is most needed in the summer and fall, but climate change has already decreased the snowpack and there is a grave concern about meeting water demand [Mote et al., 2005, Cayan et al., 2006, U.S. Global Change Research Program, 2009]. It is now accepted that groundwater, managed more than any time in the past, will play an increasingly important role in meeting needs and modulating supply in many areas across the globe, including California [Cayan et al., 2006, EPA, US EPA et al., 2011]. The management of groundwater is often plagued by policy and technology constraints. Improved management is dependent on enhanced understanding of aquifer hydraulic properties, such as permeability and storage, and supply/demand dynamics. Traditional methods of aquifer characterization, such as drilling a well and performing pump or slug tests are costly yet result in limited quantitative information that is only relevant within a small volume near the well. These methods limit our ability to understand system responses to changing dynamics (i.e., modification of pumping, alterations in recharge or river flow due to natural or anthropogenic forcing) and quantitatively characterizing large-scale aquifer hydraulic properties. Recent advances in hydraulic tomography [e.g., Cardiff and Barrash, 2011, Cardiff et al., 2013] are promising but this and other emerging characterization methods rely on expensive sensors and conventional data processing techniques. For example, each pressure transducer of high accuracy costs over one thousand dollars and typically can be

2

expected to last a couple of years. Usually these transducers are connected to a computer system using fiber optic tubes and wires, resulting in untidy jumbles of wires and the inability to operate over longer distances. Conventional methods find it hard to utilize the massive data sets that are collected. In this research, we have been making progress in taking advantage of advanced sensors, wireless networking, and smart data analysis algorithms. Pressure data over large scales can be used for many useful purposes. Here, we focus on two novel technologies, made possible only recently thanks to advances in sensor technology: oscillatory hydraulic tomography (OHT) and Ambient Signal Tomography (AST). As we are in the process of demonstrating, these approaches have the potential to provide quantitative characterization of aquifer hydraulic properties and monitor the state of aquifers at scales of interest to managers of industrial and agricultural sites, water districts, and water utilities.

PERSONNEL This research is a collaborative effort between the co-PIs, Professor Peter K. Kitanidis of Civil and Environmental Engineering and Professor Thomas Kenny of Mechanical Engineering. The following students have participated in this research: Wentao Zhang, from Civil and Environmental Engineering, Daniel Heywood, Janna Rodriguez, and James Miller from Mechanical Engineering, and Tania Bakhos from Computational and Mathematical Engineering. The wireless data acquisition system was subcontracted to Orrin Bigelow of Boise State and Hawk M2M, LLC.

PROGRESS ON MODELING We use modeling to develop numerical models to test realistic synthetic cases of aquifer and pumping conditions. This part of the research guides the sensor infrastructure prototyping and development (e.g., determine the sensitivity needed), provide relevant testing parameters and scales (for field implementation), and provide cases for further development and testing of data analytics methods for processing/imaging.

Two groundwater models are built using FEniCS (http://fenicsproject.org/), an extensive collection of modular software tools to apply finite-element analysis. The first model is at the scale one may encounter at a site remediation or a managed aquifer recharge project (100m by 100m), and the second is in a sub-basin scale (1000m by 1000m). Two different conductivity conditions are included in the model: homogeneous and heterogeneous. So in total there are 4 test aquifers.

Figure 1 Log-Conductivity Field of the Heterogeneous Model (104m2 size, the pumping well is at

the center and two observation wells are at top left) In the homogeneous case, the global conductivity is 10-5 m2/s. In the heterogeneous case, 4 square subzones are assumed to be less permeable, with the conductivity as low as 10-7 m2/s, while the remaining part is the same as that in the homogeneous case (Figure 1).

3

The boundary conditions of all models are Dirichlet (constant hydraulic head). While the conductivity is assumed different in the models, the storativity is the same for all, which is 10-5. Oscillatory pumping is included as the source/sink term in the center of the model. As Equation (1) shows, the real part of Q(t) represents the flux in the well at time t [Cardiff et al., 2013]. As time elapses, the hydraulic head among all the field will reach steady-periodic [Bakhos et al., 2014].

0( ) iwtQ t Q e (1)

Several observation points are selected to show the steady-periodic hydraulic head change (Figure 1). In order to see the difference between homogeneity and heterogeneity, one observation point is in the center of the subzone and another point is on the edge of the subzone.

Figure 2 Steady-Periodic Hydraulic Heads at different locations (104m2 size, top: homogeneous case

bottom: heterogeneous case, from left to right: field center - subzone center - subzone corner) From the periodic responses in Figure (2), it can be seen that the amplitude decreases in the further observation point from the center, because the source/sink term has less impact on these points. Moreover, fluctuations are even smaller in the heterogeneous case, given that the field is less permeable and thus less “connected”. We are in the process of conducting this modeling and we expect that we will have results within the next couple of months. PROGRESS ON PRESSURE TRANSDUCERS DESIGN We have been focusing on issues of sensor selection and waterproofing.

Bosch BMP 280 Pressure Sensor The first water pressure sensor that we investigated is the Bosch BMP 280 pressure sensor. Instead of spending several weeks to design and order a custom printed circuit board (PCB) to solder the Bosch pressure sensor package to, we decided to order a commercial PCB with pressure sensor and temperature sensor included. Details on the pressures sensor are shown here: http://www.adafruit.com/products/2651. Three of these ordered PCBs were received at our lab and have been the object of careful examination, particularly with respect to performance under water.

4

Figure 3 Bosch BMP 280 Pressure Sensor

This sensor has a maximum allowable pressure of 110 kPa absolute. With atmospheric pressure at 101 kPa, this allows this sensor to measure pressure while submerged under no more than 1 meter of water. If a greater submersion depth is desired, we will most likely need to consider pressure sensors with a higher range. However, testing this board should allow us to work out the MEMS pressure sensor water proofing strategy and put together a working code for recording the measurements. The package size of the Adafruit commercial printed circuit board is 20 mm x 20 mm x 3 mm, not including silicone water-proofing. This size can be reduced to perhaps 4 mm x 6 mm x 3 mm if we decide to design a slimmer custom PCB for the pressure sensor package. We are currently working on determining what cables can be used to run into the water tank and coat the board in silicone. We will coat one board completely in silicone, including the pressure sensor, and coat another board everywhere except the pressure sensor. EPL Surface Mount Sensor

Figure 4 EPL Surface Mount Sensor

The EPL pressure sensor is an analog instead of digital pressure; it measures changes in pressure as a change in voltage. The primary advantage of this pressure sensor is that it is available in a range beyond 7000 kPa absolute (over 700 meters of submerged depth). The device is not MEMS-based, and has a package size of 4 mm x 7 mm x 1 mm.

5

We are currently in communication with the Measurement Specialties Company that sells these pressure sensors to acquire more information on these sensors. Arduino Data Acquisition Progress of Bosch BMP 280 Sensor For initial testing purposes we plan to use an Arduino Uno microcontroller to interface with the pressure sensor and record the measured data. The Arduino Uno is a common open-source unit with wide support and a rich library of functions, including a library specifically developed to interface with the Bosch BMP280 pressure sensor. In the case of the BMP280, communication between the microcontroller and the sensor will be achieved according to the I2C protocol, a popular digital protocol that is robust and simple to implement. In addition to interfacing with the sensor, the microcontroller will also interface with a real-time clock and an SD card writer. The real-time clock will allow a time stamp to be attached to each pressure measurement. Each pressure reading, along with the time of the reading, will then be written to a removable micro-SD card as a comma-separated text file. Upon conclusion of the measurement session the card can be removed and read by a computer. The components to implement the system described have been purchased, and we are currently working to develop the necessary code. Bosch BMP 280 Testing Plan We are currently testing the devices in open air to work out the details of the measurement code, and then will be ready to test the devices while they are submerged. We suggest two possible approaches for measuring the water pressure in the tank. The first approach involves mounting the pressure sensor to the bottom of the tank and running wires up along the edge of the tank to the outside Arduino and power supply. The water level will be measured using a yardstick along one edge of the tank. The expected water pressure on the sensor can then be directly calculated from the water depth. The advantage of this approach is its simplicity in computing the water pressure from water height. The second approach would involve mounting the sensor to a 1-meter pole and running wires up to the top and off of the pole. The pole would have a yardstick along one side extending from the pressure sensor up along the pole. The sensor would be lowered vertically to a prescribed depth along one side of the tank and the pole would be fastened at that depth. The advantage of this approach would be the ease in inducing a dynamic pressure change on the sensor. By lowering or raising the sensor a prescribed distance within a prescribed time and at a constant rate, we would induce a known dynamic pressure change that could then be compared to the pressure sensor readings. In this way, we can determine the bandwidth of the sensor. Based on our group’s experience, all sensors have temperature dependence, so we also need to ascertain the dependence of this device’s pressure measurement on the temperature of the water. We therefore suggest two approaches for changing the water temperature while the sensor is submerged a fixed distance underwater. In each approach, the pressure sensor and temperature sensor takes simultaneous pressure and temperature readings over a length of time determined by the time scale of temperature change. The plots of the temperature and pressure versus time could then be compared to evaluate the cross-sensitivity of the pressure sensor to temperature changes. PROGRESS WIRELESS DATA ACQUISITION SYSTEM We have developed and tested a system that transmits wirelessly data from pressure transducers in multilevel-sampled wells to the data acquisition system, thus replacing lengthy and fragile fiber-optic cables. In April 2015, we tested this system in the field, at the Boise Hydrogeophysical Research Site (BHRS), confirming the ability to transmit analog signals reliably and accurately, from the FISO FPI-HR

6

Module to the National Instruments cRIO DAQ, using only low-cost XBee radios and circuits built on breadboards. Building on this success, a custom circuit board was then created to plug directly into the FPI-HR module to eliminate wires and evaluate a critical component of what could-be a commercial hydraulic tomography product. The XBee radios demonstrated that they are capable of replacing existing copper wires and/or fiber optic cables. In doing so, they have the ability to provide excellent wireless gateways (data transmission from one device to another device) for research-grade or professional-grade data acquisition systems. CONCLUDING This project gave the opportunity for collaborative research between the groups of Professors Kitanidis and Kenny. This research is continuing and we will have more to report within a few months.

BIBLIOGRAPHY Bakhos, T., M. Cardiff, W. Barrash, P.K. Kitanidis (2014), Data processing for oscillatory pumping tests, Journal of Hydrology, 511, 310-319. Cardiff, M. and W. Barrash (2011), 3-D transient hydraulic tomography in unconfined aquifers with fast drainage response, Water Resour. Res, 47(12), W12518. Cardiff M., T. Bakhos, P.K. Kitanidis, W. Barrash (2013), Aquifer heterogeneity characterization with oscillatory pumping: Sensitivity analysis and imaging potential, Water Resour. Res., 49, 5395-5410 Cayan, D., A. Luers, M. Hanemann, G. Franco, and B. Croes (2006), Scenarios of climate change in California: An overview., California Climate Change Center, Public Interest Energy Research Program, CEC-500-2005-186-SF, Sacramento, CA. Available at: http://www.energy.ca.gov/2005publications/CEC-500-2005-186/CEC-500-2005-186-SF.PDF Chiang, C-F., A.B. Graham, B.J. Lee, C-H Ahn, E. Ng, G.J. O’Brien and T.W. Kenny “Resonant pressure sensor with on-chip temperature and strain sensors for error correction”, Proceedings 2013 IEEE 26th International Conference on Micro Electro Mechanical Systems (MEMS), Taipei, Taiwan 45-48 (2013). Chiang, C-F. A.B. Graham, E.J. Ng, C.H. Ahn, G.J. O’Brien and T.W. Kenny, “A Novel, High-Resolution Resonant Thermometer used for Temperature Compensation of a Cofabricated Pressure Sensor”, Proceedings 2012 Solid State Sensors and Actuators Workshop, P54, Hilton Head, SC (2012). Harley, J.A. and T.W. Kenny, Design and Process Optimization of Piezoresistive Cantilevers, JMEMS 9, 226-235 (2000). Mote, P. W., A. F. Hamlet, M. P. Clark, and D. P. Lettenmaier (2005), Declining mountain snowpack in western North America, Bulletin of the American Meteorological Society, 86(1), 39-49. Partridge, A., J.K. Reynolds, B.W. Chui, E.M. Chow, A.M. Fitzgerald, L. Zhang, N.I. Maluf, and T.W. Kenny, A High-Performance Planar Piezoresistive Accelerometer JMEMS 9, 58-66 (2000). Snieder R. and K. Wapenaar (2010), Imaging with ambient noise, Physics Today, 63(9), 44 Volcano PrimeWire Prestige Pressure Guide Wire Data Sheet ©2013. Volcano Corporation, San Diego, CA. U.S. EPA, California Department of Water Resources, and U.S. Army Corps of Engineers (2011), Climate Change Handbook for Regional Water Planning. Available at: http://www.water.ca.gov/climatechange/CCHandbook.cfm U.S. Global Change Research Program (2009), Global climate change impacts in the United States, Cambridge University Press.

The Stanford Woods Institute for the Environment & Stanford Energy 3.0

Page 1

1. SEED GRANT OVERVIEW

Final Report

30 November 2015

Deployment of Soot-particulate Sensors in Flue-gas Stacks

Principal Investigator

Debbie G. Senesky, Ph.D. Assistant Professor Aeronautics and Astronautics Department Stanford University Email: dsenesky@stanford.edu

Co-Investigator

Stephen Luby, MD Professor of Medicine Infectious Diseases & Geographic Medicine Stanford University Email: sluby@stanford.edu

2. RESEARCH OBJECTIVE Pollution from brick kilns in Bangladesh, especially particulate matter <2.5 µm causes thousands of premature deaths each year. The development of robust sensing technology can aid in real-time monitoring of flue and exhaust gases, which can help to improve combustion efficiency in kilns and guide regulatory efforts and enforcement. State-of-the-art sensors based on silicon technology are limited to temperatures below 200°C and are not suitable for monitoring flue exhaust conditions. As a result, new material platforms that utilize temperature-tolerant, ceramic semiconductor materials are proposed for flue emissions monitoring. Technology developed previously by the PI, has demonstrated the chemical and mechanical robustness of wide bandgap ceramic sensors at temperatures as high as 600°C and in dry steam. In addition, the polluting methods used in the manufacture of bricks in Bangladesh have been identified by the co-PI and there is a critical need to monitor and map these harmful flue gases for extended periods. In the proposed work, a micro-scale soot-particulate sensor that can withstand the harsh flue-stack environment will be developed to provide input data to long-term pollution data profiling, which do not yet exist for soot-particulate matter at a significant scale. The specific goals of the proposed research are to (1) design and fabricate a soot-particulate sensing system, (2) perform electrical characterization of the soot-particulate sensor operation using a soot particulate generator system (3) perform 100-hour durability tests of sensor components in an actual kiln and (4) lay the groundwork for deployment of the soot-particulate sensing system within brick kilns in Bangladesh for a proceeding 1-year pilot program. These tasks aid in the realization of advanced soot-particulate sensors for deployment within brick kilns as well as other large-scale soot-generating systems (e.g., industrial power plants, automotive exhausts). Ultimately, the sensor deployment technology developed in this interdisciplinary program can lead to a reduction in harmful emissions through data acquisition and analytics. 3. MOST SIGNIFICANT TECHNICAL ACHIEVEMENT(S) • Designed and fabricated miniature soot sensor using metal-semiconductor interfaces and GaN-on-

sapphire substrates. • Generation of particulate matter 2.5 (PM2.5) using a lab-scale miniaturized chimney. This system was

used to characterize the sensor characteristic before deployment. • Demonstrated recovery of operation of the micro-fabricated soot-particulate sensor after chemical

regeneration at 550°C.

The Stanford Woods Institute for the Environment & Stanford Energy 3.0

Page 2

• Deployed sensors and sensor materials (GaN, SiC, Si and sapphire) in actual brick kilns located in Bangladesh to examine the material durability upon 100+ hours of exposure to different flue gas components.

4. ACTIVITIES AND ACCOMPLISHMENTS Outcome #1, Design and Fabrication of a Soot-particulate Sensing System: The designed soot-particulate sensor is composed of interdigitated (IDT) Pt electrodes on a GaN-on-sapphire (Al2O3) substrate to form a Schottky contact, as shown in Figure 1a. In this simple design, charged soot particles deposited on the sensor surface create electrical field effect on GaN surface and near the depletion region. Thus, particles deposited on lateral depletion region (uncovered by metal) and GaN surface affect the effective Schottky barrier height (SBH) and conduction band bending, respectively. This leads to a change in the current from the metal to the GaN substrate. Because the overall charge transport was limited by the Schottky contact, the current measurement was sensitive to change in SBH. Therefore, even a small number of soot particles can be detected by the direct deposition of soot particles near the lateral depletion region and on GaN surface, preventing the need for an initial accumulation of soot particles on the sensor surface to detect the change of resistance between electrodes by forming conductive pathways in conventional devices. To estimate the depletion width and the change of SBH in an ideal Schottky diode, the equilibrium energy band diagram, as shown in Figure 1b, can be utilized.

(a) (b) Figure 1: (a) Schematic of the Pt-GaN soot-particulate sensor and cross-sectional view of the device. (b) Energy band diagram change with respect to the positively charged soot particles on the (i) lateral depletion region and (ii) GaN surface. To microfabricate the soot-particulate sensor, a GaN-on-sapphire substrate was spin-coated with a 1 µm–thick positive photoresist at 5500 rpm for 30 s, and the sample was prebaked at 90°C for 1 min. The pattern of IDT electrodes was then exposed to UV light via a standard soft-contact photolithography aligner. Developer was used to remove the exposed IDT electrodes pattern. After rinsing the sample in deionized water, a 40 nm–thick Pt metal film was deposited on the patterned sample via the electron beam metal evaporation, and a lift-off process in acetone with sonication was performed to remove unwanted metal on the photoresist. The sample was finally annealed (preconditioned) at 500°C for 3 h before testing. To deposit soot-particulate matter on the fabricated sensor, a laboratory-scale miniaturized chimney was installed using a beaker, funnel, and compressed-air tank, as shown in Figure 2a. Commercially available paraffin candles were used to generate soot-particulate matter. The fabricated sensor was placed at the end

The Stanford Woods Institute for the Environment & Stanford Energy 3.0

Page 3

of the funnel outlet to protect the sensor from the high temperature of the candle flame (typically near 1400°C), which may induce device failure by thermal shock, degradation, or oxidation of the Pt metal. Figure 2b shows the soot-particulate sensor fabricated on 1×1 cm2 GaN-on-sapphire substrate before and after 100 µg soot accumulation.

(a) (b)

Figure 2: (a) Image of the miniaturized chimney to deposit soot particles on the sensor. (b) Image of the fabricated sensor on GaN-on-Sapphire before and after soot accumulation. Outcome #2, Electrical Characterization of Soot-particulate Sensor: Figures 3 shows the scanning electron microscope (SEM) images of the Pt IDT electrodes on the GaN layer after the deposition of 50 µg of soot particles using the miniaturized chimney system. The SEM image shows uniform and conformal accumulation of fine soot particles with diameters ranging from 30 to 50 nm was achieved along the electrodes, as shown in Figures 3b and 3c. This confirms that the size of the particles generated is within the PM2.5 particulate class, which is a main target for many soot particulate sensing applications. Individual particles were aggregated side by side, forming a chain-shaped cluster, because soot particles have both positively and negatively charged particles, which can induce aggregation of particles due to Coulomb interactions.

(a) (b) (c)

Figure 3: (a) SEM image of the interdigitated electrodes on GaN layer after soot accumulation. (b) and (c) SEM image of soot-particulate matter deposited on the sensor. To characterize the fabricated soot-particulate sensor based on a Pt-GaN Schottky interface, the current-voltage response was measured using a high-temperature probe station and semiconductor device analyzer. Figure 4 shows the change of current through the GaN layer between IDT electrodes before and after 10 µg of soot accumulation. As a result, the current of 816.2 µA at 3 V was increased to 1172 µA (i.e., ~43.6% increase) after soot accumulation, because positively charged soot particles induced conduction band bending and the SBH change at the interface of the Pt-GaN layer. Based on the

The Stanford Woods Institute for the Environment & Stanford Energy 3.0

Page 4

relationship in thermionic-emission theory and experimental data, it was found that the SBH was reduced by approximately 1.1 meV upon exposure to soot-particulate matter. The change is relatively small compared to those of gas sensors (order of ~10 meV). To demonstrate the ability of regeneration, the chuck in the probe station was heated to 550°C and held for 10 minutes to oxidize soot particles and any other contaminants on the sensor, and the sensor was then cooled down to the ambient temperature. After taking SEM images of the regenerated sensor, no soot particle was detected on the sensors, they were clear as initial. The sensing signal was measured again before and after soot accumulation. The results in Figure 4 show that the overall value of the reference current for all voltage ranges was lower than the reference current before regeneration (square-solid line), which might be caused by increased contact resistance due to high temperatures (e.g., oxidation of Pt/GaN interfaces) and tend to saturate near 550 µA at 3 V after several regeneration processes. After the second soot accumulation on the sensor, the sensing current was increased from 582.3 µA at 3 V to 892.8 µA again, showing the recovery of sensitivity, and, therefore, demonstrating the stable operation of the sensor in harsh, particulate-rich environments.

Figure 4: Measured current-voltage response of soot sensor based on Pt-GaN Schottky contact before and after regeneration at 550°C under forward bias.

Outcome #3, Deployment of Soot-particulate Sensor Materials within Brick Kilns in Bangladesh: Around 92% of the kilns for manufacturing bricks in Bangladesh are of the severely polluting, energy intensive Fixed Chimney Kiln (FCK) type (Fig. 5a). The other types such as Zigzag Kiln, Hybrid Hoffmann Kiln (HHK), Vertical Shaft Brick Kiln (VSBK) are less polluting but capital intensive, hence only comprise the remaining 8%. For this study, we deployed our sensor materials in an FCK (Figs. 5b and c) located in the Savar district, Dhaka city, Bangladesh, owned by Mr. Asadur Rahman, Vice President, Bangladesh Brick Manufacturing Owners Association (BBMOA). For effective sensing, the micro-fabricated soot sensor needs to be placed in close proximity to the flue gas vent chimneys, while keeping a safe distance to limit temperatures below 550ºC and ensure safe operation of the device. Besides particulate matter (PM2.5), the flue gases from brick kilns also contain other harmful gases (SO2, NOx) due to burning of impure coal in a combustive environment. Hence it is crucial to study the sensor material stability to these emissions. We chose five samples for each test device: (i) Si, (ii) SiC, (iii) Sapphire, (iv) GaN-on-sapphire, (v) Pt/GaN-on-sapphire (soot sensor). They are housed in a brass holder machined at the Stanford Machine Shop (Fig. 5d). Three such test devices were prepared and deployed at three different locations on the chimney between August 24-25, 2015.

0 0.5 1 1.5 2 2.5 30

0.2

0.4

0.6

0.8

1

1.2

1.4

Voltage (V)

Cur

rent

(mA)

Before soot accumulation1st soot accumulationAfter regeneration at 550°C2nd soot accumulation

The Stanford Woods Institute for the Environment & Stanford Energy 3.0

Page 5

The first device was placed at the chimney base (Figs. 5e and 5f), where the flue gases from different combustion chambers join to enter the chimney. The second and third devices were placed at heights of about 17 feet and 27 feet from the chimney base, through narrow openings made into the chimney wall which is about 20-25 inches thick. Iron rods were used to mount the test devices and provide support during operation of the kiln. Post installation, the kiln was fired on November 13, 2015, and the devices were then removed on November 21, 2015 after 8 days (192 hours) of continuous exposure to high temperatures, soot, and flue gases, as shown in Figs. 5(g-i). The exposed samples are being shipped to Stanford, where they will be further analyzed using chemical, surface, and electrical analysis to characterize material degradation and sensing response. With this knowledge, the next generation of soot sensors can be developed for enhanced robustness and stability.

(a) (b) (c)

(d) (e) (f)

(g) (h) (i)

Figure 5: (a) Schematic of a fixed chimney kiln (FCK) showing the different combustion chambers and the flue gas flow into the chimney and exit to atmosphere. [GreenTech report, 2013] (b) Image of the FCK in Savar area used for this study, showing bamboo scaffolding used to climb the chimney for deployment. (c) Close-up image of personnel deploying a device at a height of 27 ft from the base. (d) Picture of the sensor test device prepared at Stanford. (e) Picture of personnel installing a device at the

The Stanford Woods Institute for the Environment & Stanford Energy 3.0

Page 6

chimney base. (f) Picture of device at the chimney base, facing the direction of flue gases.(g-i) Pictures of device after 8 days of exposure to high temperature, soot, and flue gases.

5. FUTURE WORK/OUTLOOK FOR SOOT PARTICULATE SENSOR RESEARCH Through this seed program, we experimentally examined the electrical characteristics of soot-particulate sensors in experimental environments. In addition, we deployed actual sensors and sensor materials for feasibility/reliability tests in kilns located in Bangladesh. Upon further examination of the materials the appropriate sensor packaging and interface electronics will be determined. A future strategy for soot-particulate sensor deployment is outlined below. • Perform further sensor characterization to create detailed maps of soot-particulate concentration and

I-V characteristics at various temperatures. • Explore the use of 3D sensor architectures to increase sensor surface area and improve sensor

sensitivity. • Package sensors (soot-particulate and temperature) with power management and interface electronics

for periodic, wireless communication of sensor data. • Develop improved deployment strategies to enable sensor insertion at the top of the brick kiln and

within multiple brick kilns. For example, leverage Unmanned Ariel Vehicles (UAVs) or “Wonder Climbers” (tree climbing robots) that can readily access the brick kilns to deploy sensors and periodically collect sensor data. It should be noted that UAVs/drones are not currently permitted in Bangladesh and permission is required.

• Larger deployment effort (across South Asia) including power plants and propulsion systems (automotive and aircraft exhaust) to capture big data on the soot-particulate emissions.

Ultimately, the work spawned from this program will enable scientists to better measure emissions that have a substantial effect on the environment and health and use this information to guide approaches to reduce these impacts. 6. CONFERENCES

- Oral Presentation: Stanford Energy 3.0 Conference, “Robust Sensing Technology for Energy Applications,” Prof. Debbie G. Senesky, 14 May 2015.

- Poster Presentation: Hongyun So, Minmin Hou, Sambhav R. Jain, Jongwoo Lim, and Debbie G. Senesky, “Soot-particulate matter sensor using Pt-GaN Schottky contacts at high temperatures,” IEEE Sensors 2015, Institute of Electrical and Electronics Engineering (IEEE), Busan, Korea, 1-4 Nov 2015.

7. PUBLICATIONS

- Hongyun So, Minmin Hou, Sambhav R. Jain, Jongwoo Lim, and Debbie G. Senesky, “High-Temperature Soot-Particulate Sensors using Pt-GaN Schottky Contacts,” Applied Surface Science, under review.

- Sambhav R. Jain, Hongyun So, Debashish Biswas, Stephen Luby, and Debbie G. Senesky, “Weathering of Wide Bandgap Sensor Materials in High-temperature Particulate-rich Brick-kiln Environments,” in preparation - pending data analysis.

8. NEW TECHNOLOGY, REPORTABLE ITEMS, INVENTIONS, AND PATENTS - Debbie G. Senesky, Hongyun So, and Sambhav R. Jain, “Three-Dimensional Nanoengineered

AlGaN/GaN Textured Surfaces for Chemical Sensing Applications,” U.S. Patent Disclosure, in preparation.

The Stanford Woods Institute for the Environment & Stanford Energy 3.0

Page 7

9. POSTDOCTORAL RESEARCHER(S) / STUDENT(S) Assistance Type Number Roles / Comments

Postdoctoral 1 Dr. Hongyun So (Stanford Team), Development of soot sensor element

Graduate 1 Mr. Sambhav Jain (Stanford Team), Test planning and deployment of sensor materials in Bangladesh

Research Investigator 1 Mr. Debashish Biswas (Dhaka Team), Deployment of sensor

materials in Bangladesh

Chemical sensor project Pierre-A. GROSS, Tom LARSEN and Frederic LOIZEAU

Thomas JARAMILLO, Beth PRUITT and Denis SPITZER

Development of a micro-chemical gas sensor.

The goal is to detect trace amount of VOCs (Volatile Organic Compounds) in the gas phase with the highest possible

selectivity and/or sensitivity.

Introduction The electrochemical gas sensor project is now at a state

were the results of the first test pointed out the different

needs for improvement and optimization both of the

fabrication process and the testing. This report will focus

on this sensor, detail the obtained results for CO sensing,

and list the different feedbacks obtained. All those

information will help building a second generation of

electrochemical gas sensors.

Design and process All the tests presented in this report were performed

on the sensors obtained on the first of the three wafers

that came out of the micro-fabrication run in August.

The sensors of the two last wafers being completely

nonfunctioning. The reason of this nonfunctioning is

unclear since the process was kept similar for all wafers.

Nevertheless, the main issues being related to the

Nafion layer, it is more than probable that this is related

to too thin and non-homogeneous layer of Nafion. The

design and fabrication process of the sensor is presented

on Figure 1.

Results The experiments presented in this report were

proposed in order to confirm that the previously

observed redox peaks on figure 2, in the gas phase

experiment were indeed CO oxidation. The absence of

reference electrode in this experiment and the position

of the observed peak on the potential scale raises some

questions about what electrochemical process is in fact

occurring on the surface of the Pt electrodes. To do that

a set of simple experiments, presented in Figure 3, were

conducted. The first one is a three electrodes

experiment in H2SO4 with CO bubbling in the electrolyte.

The CO oxidation peak on Pt is located at 0.8 V vs RHE

on the anodic branch. Then, the second experiment will

be done in the same conditions (H2SO4) but in a two

electrode setup, to mimic the situation of the

experiment in the gas flow. In this second experiment,

the signal should be shifted on the potential scale, but it

will give the information of the location on the CO

oxidation peak without the reference electrode. Then

this location can be compared with the location of the

peak observed during the experiment done in the gas

Figure 1 : Design of the micro-electrochemical cell (top) and summary of its fabrication process (table).

Figure 2 : Cyclic voltammetry under different gas flows at 200 mV.s-1.

November report

phase (Figure 2). If the positions match, it means that

the observed redox peaks in the gas phase might indeed

be CO oxidation.

The first experiment is presented on Figure 4. The

obtained curve is typical for Pt in acidic medium and

exhibits the classically observed redox peaks of H

ads/des, Pt oxides formation and reduction and a

capacitive region. The black curve is obtained under N2

flow and constitutes a reference curve. No particular

signal is observed besides the normal ones for Pt. Then

CO is bubbled in the electrolyte, on cycles 37 to 40, one

can see the apparition of an oxidation peak at 0.15 V vs

MSE. This peak corresponds to the oxidation of CO in

CO2 on Pt 1, and the potential converted to the RHE scale

is ≈ 0.8 V vs RHE, confirming the nature of the oxidation

process. In addition, on can also see a small decrease in

the H ads/des peaks as the CO oxidation peak increases.

This is due to the adsorption competition between H and

CO, since they adsorb on the same sites on Pt. The

presence of this oxidation peak indicated that the sensor

is capable of detecting CO in “normal” electrochemical

conditions.

The second experiment is presented on Figure 5. It is

composed of two sets of curves. The first set (black, red,

blue and green curves) are obtained in H2SO4 in the 2

electrodes configuration, and then the purple curve is

just the cycle 8 of the Figure 2 superimposed for

interpretation. In H2SO4 in the 2 electrodes

configuration, the curve is symmetric with respect to the

0 voltage and 0 current point. This shape is expected for

this type of configuration, since each electrode is

Figure 4 : Experiment of CO detection in H2SO4 0.1 mol.L-1 in the 3 electrodes setup at 200 mV.s-1.

Figure 3 : Experiments conducted for testing the electrochemical sensors.

Figure 5 : Experiments of CO detection both in H2SO4 0.1 mol.L-1 (black, red, blue and green curves) and in gas phase (purple curve) in 2 electrodes configurations at 200 mV.s-1.

alternatively polarized positively and then negatively by

the same absolute value. The black curve is obtained

under N2 flow, and does not exhibit any particular redox

signal and the shape looks like the capacitive region

observed on Pt CVs. However, cycles 2 to 4 are obtained

under CO flow and one can see the apparition of a

symmetrical couple of peaks located at 0.2 and -0.2 V. In

the present experiment, this peak can only be CO

oxidation.

The position of the observed peaks can then be

compared to the one observed in the gas phase

experiment (purple curve) and one can see that the

position of the peaks match really well. This peak

matching constitutes a first proof that the signal

observed during the gas phase experiment might indeed

be CO oxidation in CO2 and that the micro

electrochemical sensor can indeed be used to detect CO.

Discussion Working with the electrochemical sensors for a little

more than 2 months has pointed out different problems

and questions which have to be addressed in order to

fabricate an improved second generation. Most of the

encountered issues concern the Nafion layer, which

appears to be too thin and inhomogeneous. This leads

to poor stability of the Pt electrodes sitting on top of it,

that’s why we could consider working with Nafion

membranes directly. It will be necessary to immobilize

such a membrane on a wafer in order to be able to

perform all the micro-fabrication processes, but the

membranes could be cleaned with H2O2 prior to all

remaining process and then be rehydrated with the

steam protocol.

Some new design of the Pt electrodes should also be

proposed, since the actual interdigitated seems to

induce a lot of stress in the Pt, leading to cracks which

might participate to its delamination during

experiments in the liquid phase.

Other materials for the electrodes should be

considered, since certain metals have enhance

electrocatalytic properties toward specific species, and

both electrodes could be in different metal.

Of course, the implementation of a reference electrode

into the device would add particular value, especially in

the eventuality of patent filling. Even a quasi-reference

electrode composed of a Ag layer would suffice for first

tests. Ideally the Ag layer should expend on the full size

of the device and be located underneath the Nafion if

the current type of design would be kept.

Conclusion & Outlook Some complementary experiments will be conducted

as soon as possible, and especially new experiments in

the gas phase under different conditions. In particular,

obtaining a curve of the sensor under hydrated N2 flow

will definitely confirm the identity of the peak observed

on Figure 2.

In parallel, calibration experiments for the capacitor

sensor must be made prior to any testing in CO

detection. In particular, the heater must be calibrated

against the applied voltage in order to be able to do CO

detection experiments at different temperatures. To do

that, an infrared laser thermometer will be used.

References (1) Vidaković, T.; Christov, M.; Sundmacher, K.

Electrochimica Acta 2007, 52 (18), 5606–5613.

Using Consumer Cameras and Crowdsourcing Data Collection to Monitor Environmental Challenges

Final Report

Henryk Blasinski, Joyce Farrell, Rob Dunbar and Brian Wandell The purpose of this grant was to build algorithms and devices to assess the health of coral reefs. The methods were designed to enable citizen scientists to contribute their data to help monitor the health of the oceans. During this year we accomplished the following main objectives. First, we developed a computational imaging method that includes both algorithms and hardware for assessing the spectral reflectance and fluorescence of coral reefs. Second, we implemented a hardware system to make this measurement. The system comprises a water­tight housing, considerable electronics including a digital camera, and a spectrally controlled light source. This system is being installed by Stephen Palumbi at Hopkins Marine Station, and an experiment measuring the spectral signature from coral reefs under controlled temperature conditions is under way. Third, we created a website that permits people to upload images and have them processed by the methods described in our algorithms. This site is preliminary, but demonstrates the principles and a first pass of how the data will be crowd­sourced and analyzed. Finally, building on the Woods Institute support, we reached out to journalists to seek funding to deploy this citizen science project at a slightly larger scale in a project supported by the Brown Institute for Media Innovation.

Reflectance and fluorescence modeling in the underwater environment The amount of coral fluorescence is correlated with the overall health the reefs [5,6]. Specifically, corals that are adversely affected by changes in the environment, such as salinity, water temperature or nutrient contents, exhibit lower concentrations of fluorescent molecules. For example an increase in water temperature leading to coral bleaching first affects corral fluorescence properties [7]. These changes in fluorescence are hard to observe under normal daylight conditions.

We developed a computational imaging system that simultaneously estimates surface spectral reflectance and fluorescence properties in an underwater environment and daylight illumination [1,2]. The system hardware is composed of a conventional RGB camera and a spectrally controlled light source built using a small number of narrowband LEDs. The system computation separates surface reflectance and fluorescence properties from multiple images of the scene, each illuminated by a different LED. The computation uses a linear image formation model to express the measured pixel intensities in terms of the reflectance and fluorescence properties and then solves an inverse estimation problem using convex optimization techniques.

Modeling underwater imaging is challenging because water affects the way light propagates between the light source, scene and the camera. Specifically, water attenuates certain light wavelengths more strongly than others and particles in the water increase backscatter

contribution. This means that the appearance of an underwater target can vary substantially, depending on the specific properties of the medium. We developed and implemented an underwater imaging simulation toolbox that relates different optical phenomena to a small number of independent, biologically meaningful parameters, such as phytoplankton or colored dissolved organic matter concentrations [3]. The simulation environment is helpful for optimizing algorithms and camera systems for underwater imaging applications.

Accomplishments: The computational ideas were presented at the Optical Society meeting in Arlington, Virginia. The specific algorithm is described in a paper that is under review at IEEE Transactions on Pattern Analysis and Machine Intelligence. The source code implementing these methods is available at a github repository that we will open­source once the paper is published.

Active illumination underwater imaging system In order to perform the separation between reflectance and fluorescence components it is necessary to capture color images of a target illuminated with lights having different spectral power distribution. Existing underwater cameras can control external light strobes, but none of those strobes provide light that can be spectrally adjusted. To overcome these limitations we designed and built a custom imaging system composed of off­the­shelf components: seven narrowband LEDs, a color camera and a Raspberry Pi minicomputer synchronizing lights with image acquisition. All components were placed inside a water­tight housing.

The system can be operated in two modes. In the fully autonomous mode the LEDs and the camera are powered from an internal battery pack giving a few hours of operation. Alternatively, the system can be connected to an external power supply and semi­submerged, allowing observations in shallow water tanks. In this mode the camera becomes less versatile, but it can be used to monitor changes in coral appearance over extended periods of time. This configuration is the most applicable to controlled laboratory experiments, where environmental changes can be correlated to measured pixel intensities.

Accomplishments: A reflectance and fluorescence imaging system. This system is being installed by Stephen Palumbi at Hopkins Marine Station, and an experiment measuring the spectral signature from coral reefs under controlled temperature conditions is beginning. Our system is much less expensive and more portable than previous imaging systems that were designed to measure coral reef fluorescence in situ [8,9]. Our analytical methods also make it

possible to 1) simultaneously measure and separate the reflectance and fluorescent components in coral reef images and 2) measure multiple and unknown fluorophores.

Data crowdsourcing We investigated the challenges and opportunities for crowdsourcing coral reef image data. The idea is to leverage the increased popularity and affordability of underwater photography and engage the large population of amateur divers who capture pictures for pleasure during their trips. We propose to provide divers with our custom LED light source connected to a standard action camera such as GoPro and ask them to capture images of coral reefs during their trips. After the trip, participants would upload their geo­tagged images to a web­based repository where the data are analyzed and coral health evaluated by extracting the fluorescence signal intensity. Community engagement would build a global dataset of coral reef images bolstering monitoring efforts.

To illustrate the idea we created a simple website where coral images can be uploaded and analyzed. Currently the site processes images of different types of coral obtained by simulating the active illumination acquisition system. Uploaded images are analyzed with the reflectance­fluorescence separation algorithm and the reflectance spectrum as well as fluorescence excitation­emission properties are displayed to the user.

Accomplishments: A simple website for coral image analysis (http://reef.stanford.edu, user: reef, password: r33f).

Future work This project has paved the way for two directions of future research. First, by installing our imaging system in Prof. Stephen Palumbi’s lab at the Hopkins Marine Station, we can facilitate additional controlled studies of coral reef health as a function of environmental conditions (temperature and salinity). Images captured by our system will quantify changes in reflectance and fluorescence in response to different environmental stimuli. The experimental dataset will be used to derive statistical models relating camera pixel intensities with the state of coral reefs.

Second, we envision carrying out a small scale data for the crowdsourcing study. A selected group of divers, (participants in the Stanford at Sea program) could be provided with spectral light sources and asked to capture pictures during their dives. They will be asked to contribute those pictures to the online repository. Success of this small scale experiment would pave the way for a global coral reef monitoring effort.

References

1. H. Blasinski, J. Farrell and B. Wandell; ‘Simultaneous Surface Reflectance and Fluorescence Spectra Estimation,’ IEEE Transactions on Pattern Analysis and Machine Intelligence, TPAMI (under review)

2. H. Blasinski, J. Farrell; ‘Simulation of Underwater Imaging Systems,’ Imaging Systems and Applications 2015, Arlington, VA

3. H. Blasinski, J. Farrell; ‘A Three Parameter Underwater Image Formation Model,’ Electronic Imaging 2016 (accepted for publication)

4. E. Fuchs ‘Separating the fluorescence and reflectance components of coral spectra,’ Applied Optics 40(21), 3614­3621, 2001

5. M. Warner, W. Fitt, G. Schmidt; ‘The effects of elevated temperature on the photosynthetic efficiency of zooxanthellae in hospite from four different species of reef coral: a novel approach,’ Plant, Cell and Environment 19, 291­299, 1996

6. G. Piniak, N. Fogarty, C. Addison, W. Kenworthy; ‘Fluorescence census techniques for coral recruits,’ Coral Reefs 24, 496­500, 2005

7. R. Rowan; ‘Coral bleaching: Thermal adaptation in reef coral symbionts,’ Nature 430, 724, 2004

8. M. S. Roth, M. I. Latz, R. Goericke, D. D. Deheyn; ‘Green fluorescent protein regulation in the coral Acropora yongei during photoacclimation.’ J. Exp. Biol. 213, 3644–3655, 2010.

9. M. S. Roth, D. D. Deheyn. ‘Effects of cold stress and heat stress on coral fluorescence in reef­building corals’. Scientific Reports, 2013