The inspector sits at the controls, eyes glued to the screen as the autonomous drone flies past working flare stacks and heated gas plumes. It buzzes from place to place, identifying corrosive spots, marking them for maintenance.
If this sounds like a video game, that’s not surprising: the technology making this flight possible was originally designed for gaming. Through a new partnership between Avitas Systems, an inspection business launched by GE Ventures, and NVIDIA, a leader in artificial intelligence, the tools that make video games so immersive will help inspectors ensure the safety at places like refineries and power plants.
Industrial inspection can be a high-risk and expensive business because of the manual nature of the work. It often requires humans to scale up to the tops of flare stacks or dive to the depths of underwater pipelines. Flare stacks, for example, burn off waste gas and must be shut down for days before they become cool enough for an inspector to approach. “Many companies spend $100 million dollars plus on inspection of their industrial assets,” says Alex Tepper, managing director at Avitas Systems.
Drones or robots, on the other hand, can get close to high-temperature flare stacks and go into vapor-filled fuel tanks while they’re in operation. They can also gather high quality video, photographic images, infrared, and temperature data.
But making these drone inspections work requires processing large amounts of data — and that’s where NVIDIA comes in. “When you’re driving in a video game and you see a lifelike puddle picking up the light, it’s our technology making that imagination possible,” says Jim McHugh, General Manager of NVIDIA. “And that same underlying technology is incredibly well-suited for artificial intelligence: both take an incredible amount of computational power. By applying that power to data, Avitas Systems is able to unleash artificial intelligence to solve risk based inspection with incredible accuracy.”
During inspections, AI can help by designing specific paths to limit drone flight time, making those inspections more efficient. For example, if there are 15 points that need to be checked, the drone can travel the fastest path between those points.
During those inspections, the drones record hundreds of hours of video, produce thousands of images, and measure a host of key metrics, from temperature to metal corrosion. Added in with operational data from a plant and other information gathered from inspectors with handheld sensors, this data is processed through Predix, GE’s cloud platform for the Industrial Internet. “In Predix, we have a second level of artificial intelligence that analyzes the fused data from multiple sources, so we can understand an asset from multiple perspectives,” Tepper says.
“In Predix, we have a second level of artificial intelligence that analyzes the fused data from multiple sources, so we can understand an asset from multiple perspectives,” says Avitas Systems’s Tepper. Image credit: Avitas Systems.
But, even with Predix, this requires a massive amount of computational power. The NVIDIA DGX system, which McHugh calls “a supercomputer in a box,” allows a new level of computing power and simplicity. “If you can think about one server that is the equivalent of pretty much a row of servers, it’s that level of processing power for deep learning and AI,” McHugh says. And, unlike a row of servers, DGX can be set up in two hours — as opposed to two weeks.
Using this AI-enabled analysis, inspectors can understand industrial installations at a level that was never before possible. “We can build a 3D model of an asset and track changes over time,” Tepper says. “We can use predictive modeling, so if there’s a worrisome spot of corrosion, we can predict that it will need to be addressed before three years have passed, because we know that three years from now it will be a very big problem.”
The next step: semi-autonomous drones that think for themselves. Traditionally, data from an inspection goes to a data center for analysis. But the compact size of NVIDIA’s technology makes it possible to analyze some data much closer to the drone. “We have AI that sits on a drone or robot crawler, which allows it to react to a given situation,” Tepper says.
Ultimately, this capability will transform inspections. “NVIDIA is excited about it because it’s a new industrial application for artificial intelligence,” Tepper says. “And we’re excited because we’re getting a clearer picture of the condition of these assets, which will make them safer and better maintained.”
Top and above: “We can use predictive modeling, so if there’s a worrisome spot of corrosion, we can predict that it will need to be addressed before three years have passed, because we know that three years from now it will be a very big problem,” says Avitas Systems’s Tepper. Top GIF credit: Avitas Systems. Above image credit: Sarcos.