By Debra Werner,
Published by SpaceNews, 18 October 2023
MOUNTAIN VIEW, Calif. – Los Angeles startup Machina Labs is expanding its role in the space sector by working with satellite and hypersonic vehicle manufacturers.
Since Machina Labs was founded in 2019, the company has worked with NASA, the U.S. Air Force Research Laboratory and SpaceX to apply robotics and artificial intelligence to space-related manufacturing processes.
Work with AFRL focused on developing robotic technology for manufacturing metal tooling for composite structures. For NASA, Machina Labs developed machine learning-based software for in-space manufacturing with autonomous articulated robots.
Now, Machina Labs is working with satellite manufacturers to help them rapidly iterate designs.
“I can try out a design of a tank and then see if it works,” Edward Mehr, Machina Labs CEO and co-founder, told SpaceNews at Satellite Innovation 2023. In addition, Machina Labs helps customers manufacture “things that were traditionally not possible,” he added.
For example, doughnut-shaped toroidal propellant tanks, popular decades ago, have rarely been produced in recent years because they are difficult and time-consuming to manufacture. Machina Labs’ process, called Roboforming, reduces the cost and speeds up manufacturing of toroidal tanks, said Mehr, a former Relativity program manager and SpaceX software engineer.
For hypersonic vehicles, Machina Labs works with materials tough enough to withstand the heat of reentry.
“With our technology, we can process some of those materials like titanium or Inconel,” Mehr said.
AI and ML
Machina Labs machines, which use Nvidia chips, rely on machine learning to replicate the work of people who incrementally deform metals or composites to create shapes.
“We need to replicate what happens in the mind of a craftsman,” Mehr said.
To do that, the company built empirical models of how materials deform throughout the shaping process. Then, Machina Labs engineers determine the appropriate set of processes.
“For every geometry, what are the right process parameters and where does the robot need to go to get to the right part?” Mehr asked.
See: Original Article