Current Issue
This Month's Print Issue

Follow Fast Company

We’ll come to you.

2 minute read


Using Big Math To Help Drones Anticipate Dangerous Infrastructure Problems

A Massachusetts-based startup's analytics software is being honed for drones, and the end results could reveal if drones are viable for industrial uses like doing maintenance on power lines.

[Image: Flickr user Frederic]

The Internet of things is everywhere. Internet-connected sensors are becoming omnipresent and cheap, and manufacturers are embedding them in every single little thing. Sensors are in our televisions, our cars, and even our thermostats. Soon enough, they'll even be in unmanned aircraft flying over our heads.

Colin Gounden is the head of Via Science, a self-described "big math" company headquartered in Cambridge, Mass. Via's primary platform is a software platform called REFS that allows clients in finance, health care, and other industries to make predictions from massive data sets. In recent months, they also experimented with using REFS to make predictive analytics from UAVs.

In one of Via's newest partnerships, the company began applying their predictive analytics to input from UAV sensors. The REFS platform takes massive intakes of data sets in real time, and then uses those to identify patterns and anomalies that don't fit in within the data set. Drones, which fly autonomously in the air, are usually outfitted with an array of sensors that register visual information and more. Cameras are a sort of sensor, and Gounden's company is using UAVs to see if predictive analytics can be made from visual information.

To hear Gounden explain it, it's a dual-use experiment to see if drones for industrial use are viable and if they can predict things through their camera. Working with an MIT-associated UAV manufacturer that he did not identify, Via's software was installed on laptops connected to the company's drones. This particular firm manufactures UAVs for monitoring of sensitive infrastructure in situations that are dangerous for human pilots.

"In 2013's federal budget, there were billions in the federal budget requisitioned for UAVs. This is already a big market, and the industrial drones section of it is just getting started. The thought process we had is that we already fly helicopters out to do dangerous surveys, and we wondered if you can use a UAV instead to inspect pipeline leaks or power lines from the outside? The United States regulates 100% of all power lines each year, but only 20% get inspected because there is not enough bandwidth and ability to do it right now. We wanted to see if drones could do video surveillance for that, which is very possible," Gounden told Fast Company.

Teaming up with the UAV manufacturer, Via uses REFS to turn video feeds into databases of structured features like size, shape, color, direction, speed, time, and location. These are then used for predictive analytics about problems with the power line or nearby objects. Interestingly, Gounden says the process results in a situation where "you have lots of rows, but very few columns from machine-generated data." By collecting lots of data on a few very specific subjects, Via is able to make predictions on behalf of the UAV's operator.

The end goal? Predicting errors or broken infrastructure through the drone's cameras and other sensors. As Gounden puts it, "no one wants things to fail. People want a ROI that's very clean—and there's nothing more clear than preventing failure."