Students design IoT system for farmers
Article By : Martin Rowe
MIT undergraduates took their berrySmart device to the finals of the Keysight IoT Innovation Challenge. Here's how they developed the design and how they overcame obstacles.
On September 21, 2019, Keysight Technologies held the finals of its IoT Innovation Challenge in New York City. As one of the judges, I had the opportunity to see and hear firsthand from the six finalists as they presented their designs in the “Smart Land” and “Smart Water” categories. One of the finalists for Smart Land was a team of MIT undergraduates: Nikhil Murthy, Gabriella Garcia, Sunny Tran, and Irin Ghosh. To learn more about their design goals and challenges, I met with Murthy, Garcia, and Tran (Figure 1) near the MIT campus a month after the competition.
Figure 1 Students (l-r) Nikhil Murthy, Gabriella Garcia, and Sunny Tran on the MIT campus in Cambridge, Mass.
The team’s entry, called “berrySmart,” started as a class project, after which the students entered it in the Keysight competition. “We were taking an IoT class and berrySmart was our class project,” said Murthy. “Only later did we submit it to the Keysight competition.”
How did they come up with the concept? As Murthy explained, “After hours of brainstorming for an idea, I remembered speaking to a blueberry farmer for an economics class. The farmer didn’t have the resources that larger farms have. We then started thinking about how to use what we learned in the IoT class that could help.”
“We had heard about agricultural IoT projects,” added Garcia, “but most were robots. We wanted a project that was more software based using data collected from sensors.” What’s truly impressive is that the students had just three weeks and $80 to develop a concept that they could show to the class of 200.
Prototype
With berrySmart, farmers can set up a Wi-Fi mesh network where sensor nodes collect data on temperature, humidity, and soil conditions. A prototype module (Figure 2), shown at the competition finals, uses a microcontroller, a solar panel, batteries, and a Wi-Fi module. Only one berrySmart node (the “edge node”) needs to connect to the internet. Each node relays data to other nodes, which ultimately relay the data to the edge node.
Figure 2 The berrySmart collects sensor data and transmits it to a server over a Wi-Fi mesh network. A solar panel charges the node’s battery.
Engineering projects are full of technical decisions. Initially, the students considered developing a Bluetooth-based mesh network. After spending half of their three-week time allotment, the team realized that Bluetooth wouldn’t work because of its short range. Thus, they switched to Wi-Fi. The Bluetooth-based network was not reliable,” said Tran. “It dropped too many packets. With Wi-Fi, we saw practically no dropped packets and connections remained stable.”
“Because of the time spent on Bluetooth,” added Murthy, “we were scared that we’d never complete the project on time. We were relieved that we could use Wi-Fi and move on.”
Garcia noted that there’s relatively little online documentation on how to implement a mesh network using their microcontroller. Range was an issue, which is why they settled on five nodes per acre. “With better electronics, we can probably increase the range and use fewer nodes where needed. It depends on the crop. For a standardized crop such as corn, a farmer could use fewer than five nodes per acre.”
When asked about the sensors, Murthy explained that they use a moisture sensor in the ground, an ambient-light sensor (photoresistor), and a temperature/RH sensor. “Because the berrySmart is flexible and open source, farmers can add sensors such as pH. The mesh network is also open source. We’ve developed a software dashboard that can make recommendations based on sensor data.”
“Farmers want to have options to customize their sensors,” said Garcia. “We used a photoresistor to sense light, but the farmers expressed interest in using other sensors as well. Even so, they don’t want to build the sensor networks themselves.” Murthy noted that the farmers wanted a plug-and-play system where they don’t have to deal with wires. They want a system that can automatically detect the sensor type and collect data. Sensors such as those used for soil moistures have analog outputs while others are digital.
The electronics behind berrySmart consists of:
- ESP32 Wi-Fi transceiver
- Arduino microcontroller board with 32 channels of I/O
- Power management board
- Solar panel
- Battery
Given the three-week timeframe, the team had to use readily available components. Tran noted that a future design might use a Microchip ATTiny85 processor, which should result in lower power consumption, in part because it has fewer I/O ports. “We only needed five or six I/O ports,” said Murthy, “note the 32 on the Arduino board.” Using open-source software, Garcia designed a single-sided PCB to integrate the entire system, then the team added wires to complete the connections.
The berrySmart needs software to control each node, communicate with the network, collect data, and present the data to a farmer. They had to develop a protocol for the notes to communicate over Wi-Fi. They created an open-source library for Arduino to create the mesh network. It’s based on the HELLO protocol that established links between neighbor nodes. They also added data buffers for the Arduino. Now, anyone can use the library, as they did in this application. With the protocol, nodes can relay data to the edge node, which connects to a server over the internet. That single connection could occur at the farmhouse or at a greenhouse. “Because many greenhouses have Wi-Fi,” said Garcia, “we could use fewer nodes to get the data online. Only the smallest farms may lack Wi-Fi at the greenhouse. We worked with one small organic farm that didn’t have Wi-Fi at the greenhouse, but the farm was just five acres in size.”
At the present time, the mesh network connects to a server at MIT that hosts the user dashboard (Figure 3). Through a browser, a farmer can get a map of the farm and view sensor data. A machine-learning algorithm can inform the farmer which areas of the farm require attention and suggest remedies such as watering or adding organic matter to a given location. “Irin (Ghosh) was the web developer while I helped work on the analytics,” said Garcia. “We went through many iterations during development.” Murthy noted that giving the farmer a visual presentation, as opposed to just the sensor data, greatly enhanced the berrySmart’s value to the farmer. “Having specific recommendations was the most useful,” he added.
Figure 3 The dashboard provides farmers with plots based on sensor data, here showing data from light sensors when day turns to night.
In addition to learning about networks and protocols, the students had to learn about agriculture. For example, they had to research what the system should recommend to farmers on what to do under certain conditions. Garcia explained that, for example, conditions of high temperature might require misting crops.
The machine-learning algorithm’s support-vector machine looks at the data, which includes a geographic location, to place the sensor onto a plot of the farm. Using that positional data, the software provides an “aerial view” of the farm’s sensor nodes. It highlights extreme conditions that might require a farmer’s attention. Once such application, according to Garcia, was for sensing the temperature of raspberries, with prefer temperatures between 70°F and 80°F (21°C to 27°C). When the temperature reached 90°F (32°C), the farmer might spread mist over the plants. In cases such as this, the software doesn’t tell the farmer what to do, but simply points to areas that need attention. The recommendation based on research might apply to conditions unfamiliar to a farmer.
The berrySmart is very much a work in progress. Indeed, Murthy noted that they have secured funding for additional work. Some funds might be the result of the team’s prize of $25,000 in cash plus $25,000 in Keysight equipment for MIT. Although the team didn’t win the grand prize ($50,000 in cash plus $50,000 in equipment) nor the prize for best Smart Land project, they did receive the Keysight Diversity in Tech Award, selected by Keysight CMO Marie Hattar. “It gives me so much hope to know that these young students are our future. It’s one thing to talk about bettering our world, but they’re getting out there and doing it. By leveraging emerging technologies, coupled with an inspiring level of determination and grit, this team is primed to make a real impact. Keysight was thrilled to recognize this team with the Diversity in Tech award and I look forward to following their journey.”
When asked about the test equipment, the MIT students replied that they had selected an oscilloscope for their professor. “He was very happy,” they explained.
- IoT Student Challenge: We Have Winners
- Sensor basics: Types, functions and applications
- Sensors Design Ideas
- What are the most useful sensors?
- Temperature sensors are improving, but select carefully
- Signal Chain Basics #86: Fundamentals of Temp Sensors
- Making sense of light sensors
- Simple night-light uses a photoresistor to detect dusk