Video of man working with robots

Legs, Wings, and a Steady Hand: How the ARC Is Building Next-Generation Inspection Tools

By Brendan Coffey

Every generator, turbine blade, and transformer made by GE Vernova shares one crucial requirement: meticulous inspection. Pre-delivery and field inspection jobs are among the energy industry’s most essential tasks. But they’re demanding, often requiring weeks of planning, exposure to heights and elements, hours of painstaking visual and physical work, and costly customer downtime.

To streamline this process, the GE Vernova Advanced Research Center (ARC) in Niskayuna, New York, is rethinking how to automate. Unlike automakers that can fund custom robots for millions of identical units, GE Vernova produces high-complexity items in low volumes—sometimes making as few as eight specialized gas turbine parts a year. “A lot of our thought process is around how do we automate automation,” explains Jaron Kuppers, technology manager for robotics and computer vision, focusing on adapting off-the-shelf components for novel uses.

A new GE Vernova film offers a stylized look at the answers taking shape: a robotic dog, an inspection drone, and a camera-equipped robotic arm, all guided by AI and with the goal of eventual deployment in the field. One key aim: The learning ability of AI means the techniques developed for each tool in the near term should be flexible enough to apply to other applications down the line, creating cost-effective approaches to improving safety, quality, delivery, and cost.

Here is a closer look at four innovations the Advanced Research Center is developing to create the future of inspection:

Field engineers using a drone

Noah Wiederhold, one of the ARC’s robotics and AI engineers, is helping develop drones that can do reconnaissance for the technician and even handle some of the inspection tasks.

Image credit: GE Vernova

Taking Flight Inside a Wind Turbine

Wind turbines can be among the toughest assets to inspect. For safety, whenever a person is within 300 feet of a turbine, it has to be shut off. Then a technician climbs a ladder inside the tower, past platforms spaced vertically 60 to 120 feet apart, checking power cables along the way to the nacelle, where the gearbox and generator sit. Then a technician needs to go outside and inspect turbine blades, often rappelling down to look at blades for signs of lightning strikes and other issues.

Noah Wiederhold, one of the ARC’s robotics and AI engineers, is helping develop drones that can do reconnaissance for the technician and even handle some of the inspection tasks. “We can have human technicians better understand the environment they’re getting into when they go to perform maintenance and also spend less time up-tower, working at heights in a potentially dangerous environment.”

Using specialized confined-space drones, Wiederhold and his team are developing a pilotless drone that can navigate the turbine not only outside but inside — no small task, given the signal-scrambling effect of the steel support structure. Eventually, using AI, Wiederhold expects the drones to be able to plan their own routes and do interpretation of the thousands of photos it takes, drawing on product manuals and training images
“Automated assessment using AI and other techniques is [a] huge savings for everybody involved,” Wiederhold explains.

The Robotic Arm That Plans Its Own Photo Shoot

Every GE Vernova part gets inspected when it’s manufactured and again during regular service. Robotics already help engineers automate that work, but programming robots to inspect complex parts with them means literally weeks of back-and-forth between an inspector and a roboticist to define the camera angles needed to photograph parts and program the robot. Given the low-volume production of thousands of parts in GE Vernova products, there hasn’t been a way to effectively streamline the process.

Senior Robotics and Autonomous Systems Engineer William Tan leads a team seeking to use digital twins — a virtual model that acts like the real thing — and AI to develop a next-gen robotic arm that can automatically propose camera viewpoints based on inspector preferences. After using AI to teach the arm what it should do based on those preferences, Tan says, the movements will be tested in the digital twin environment before being deployed in real-life situations. Then an image analytics team will use AI to help root out defects and generate inspection reports. A prototype of the system is headed first to inspect turbine buckets and nozzles at Gas Power sites in Hungary, Singapore, and South Carolina, with wider deployment expected by year’s end.

Robotic Arm

“We are pushing hard on automating the whole process, from the start to the end,” Tan says. “We want to reduce it from weeks to days — even hours.”

Robotic dog that hunts

Robotic dogs could use AI to learn the product they are inspecting to streamline human programming ahead of deployment.

Image credit: GE Vernova

A Dog That Reads Before It Hunts

Commercial inspection drones often have two severe limitations: They’re too small and light to carry much battery, meaning their flight time maxes out at around 15 minutes, and they have little ability to manipulate anything in their surroundings — to open a door, for example. That provides a real purpose for robotic dogs, four-legged machines that can bear the weight of larger battery packs and also carry tools, such as a back-mounted arm, to improve inspections and even do controlled functions — say, swapping out a fuse.

Like the wind turbine drones, these dogs could use AI to learn the product they are inspecting to streamline human programming ahead of deployment. Today, according to Kuppers, the ARC team is teaching the robots to navigate flat spaces before working up to more complex environs and then, eventually, real-life assignments, like potentially inspecting a direct air capture (DAC). At the risk of personifying these tools, it’s worth noting that the dog and the drone aren’t rivals: Kuppers’s group has already designed solutions where the two work in tandem on a single task.

Exterior of building with people walking near it

A Robot That Gets a Grip

Think of Star Wars, where Luke was always minding the task droids on his family’s desert farm, each bot going about its own very specific task. Not every robot needs to be able to do multiple things: Sometimes the handiest robots could be those that are able to do a specific assignment, one that so far only humans can effectively do — like tightening a screw. For humans, it’s such an easy task that we’re hardly aware of all the inputs our body balances to know if it’s threaded correctly and just how tight is enough and what is too much. Robots can’t do this.

ARC researchers are working on developing robots that could check whether bolts are properly torqued and tighten them without overdoing it. It’s not an insignificant endeavor: It might take a technician three and a half hours to retorque the massive bolts on a wind turbine base with a 35-pound wrench. Freeing a person from that back-stressing, time-consuming work would “let technicians do the higher-value tasks,” Kuppers explains.

It’s a fitting summary of the ARC’s efforts in the future of robotics: Let machines do the dull, dangerous, and repetitive work, so people can focus on what they do best.

Want to turn breakthrough ideas into real-world impact?