When NASA’s Mars rover, Perseverance, needs to move from point A to point B, it’s not as simple as plugging in GPS directions. The rover has to avoid boulder fields, sand dunes, and steep slopes on a foreign planet.
To help chart a course for Perseverance, NASA has enlisted the help of AI and a digital twin. Digital twins are particularly useful to NASA, which “operates in some of the most extreme environments imaginable,” said Kevin Murphy, acting chief artificial intelligence officer at NASA. The technology creates a virtual replica of the actual environment and conditions, helping NASA scientists understand real-time conditions in places like outer space and Mars.
Like NASA, the broader aerospace industry is also embracing the use of digital twins with AI. Human oversight and verification are essential for aircraft and weapons that may put people’s lives at risk, said Karen Willcox, director of the Oden Institute for Computational Engineering and Sciences and professor of aerospace engineering and engineering mechanics at the University of Texas at Austin.
“So much of the AI deployment that we’ve seen in other settings, the risks of being wrong are not life and death, and that lets companies move much faster,” Willcox said.
“AI is very powerful,” she added. “But AI by itself will never be enough.”
Inside NASA’s digital twins
Digital twins aren’t exactly new to NASA. The space agency pioneered the concept in the 1960s during the Apollo mission.
Since then, the technology and its use cases have advanced. Murphy said that with AI layered on, digital twins go beyond virtual clones and can now make predictions, diagnose issues, and recommend actions in real time. AI offers more insights than digital twins alone, he added, and it can sort through sensor data coming into the digital twin faster than humans, spotting anomalies or risks that could have been missed.
The two technologies were used in tandem when NASA developed the path for Perseverance. Murphy told Business Insider that engineers at NASA’s Jet Propulsion Laboratory in California fed information to an AI model — the same kind of data they would give a human devising the Mars rover’s path. The AI generated a route that avoided dangerous terrain. NASA engineers reviewed, refined, and cross-checked against a virtual replica of Perseverance’s surroundings before sending commands to the rover.
NASA also used a set of digital twins to test and monitor its James Webb Space Telescope — an enormous instrument four stories tall and the length of a tennis court.
“It was too large to test in our thermal vacuum chamber,” Murphy said.
One twin was a 3D video-based model, which let scientists track the unfurling of the telescope’s sunshield, which is “a complex maneuver with 344 potential points of failure,” Murphy said.
The other twin modeled the telescope’s core to track its temperature. If it overheated, it could make the telescope blind and unable to observe galaxies, Murphy explained.
Julie Van Campen, a Webb telescope mission systems engineer at NASA, said they developed the twins starting in the early 2000s, before AI as we know it today existed. But the modelers who created the twins are now applying their expertise to the next generation of NASA projects, she said, which includes AI tools.
Today, NASA uses AI to analyze the enormous volume of data which the James Webb Space Telescope collects each day. With AI’s assistance, NASA can connect data sets from various observatories, helping it gain a broader perspective on the universe.
Van Campen likened the data sets to a gold mine — and AI helps unearth the nuggets and treasures buried inside.
Infinite possibilities, sky-high stakes
While enormous telescopes and Mars rovers may seem niche to NASA, Murphy sees potential far beyond space exploration. AI isn’t a single technology, nor is it a “one-size-fits-all solution,” he said.
“The downstream applications for the broader aerospace industry are endless,” he said.
Willcox said predictive maintenance is among the most common applications in the aerospace industry. Sensors stream real-time data from an aircraft or engine into the digital twin, AI updates the replica’s state in real time, and the twin can generate predictions for humans to evaluate.
This shifts maintenance from a set schedule to one where parts are replaced only when the model shows wear and tear or performance issues, and manufacturers can plan in advance for downtime. Airbus is one company that uses digital twins and AI for predictive maintenance, as well as product development and manufacturing.
Meanwhile, Boeing implements both AI and digital twin technology to simulate test conditions for its aircraft.
Willcox said aerospace companies and agencies are keen to use digital twins during testing and evaluation, which can be time-consuming and costly. In developing a military fighter jet, for example, a quarter of the budget can go toward test and evaluation, she said. The Defense Advanced Research Projects Agency is running a research program around AI and digital twins to accelerate the testing of combat systems — all while maintaining safety.
But Willcox said the risks are high in the aerospace industry, requiring greater verification and validation of AI outputs than in other industries.
“You could go much faster if you didn’t have to worry about being wrong,” Willcox said. “That’s not the way we do things in aerospace.”
She sees AI and digital twins complementing humans, where the technology provides a two-way flow of information between physical and virtual environments, allowing professionals to interact with the tech in natural language and make decisions in real time.
Murphy said NASA’s teams conducted “extensive testing” before launching AI technologies with the Mars rover. He added that they checked more than 500,000 variables before sending commands to Perseverance. And they continuously verify and update digital twins against real-world performance.
It all comes back to a central mission: explore options, make decisions, and spot problems — “without putting people or hardware at risk,” Murphy said.
Read the full article here
