Business

BMW’s Virtual Factory Uses AI to Hone the Assembly Line

April 12, 2021

German carmaker BMW plans to begin making drivetrains for electrical automobiles at an unlimited manufacturing unit in Regensburg, Bavaria, later in 2021. Well earlier than any new elements roll off the manufacturing line, the total manufacturing course of will run in stunningly practical element inside a digital model of the manufacturing unit.

The simulation permits managers to plan the manufacturing course of in larger element than was beforehand potential, says Markus Grüeneisl, who leads manufacturing technique at BMW. “We now have a perfect digital twin of our real-time production,” he says.

The simulation is a part of BMW’s plan to use extra artificial intelligence in manufacturing. Grüeneisl says machine-learning algorithms can simulate robots performing advanced maneuvers to discover the best course of. Over time, BMW needs to use the simulation to have robots find out how to carry out more and more advanced jobs.

BMW used a software program platform known as Omniverse, developed by the chipmaker Nvidia, to recreate the Regensburg manufacturing line. Last 12 months, BMW stated it was utilizing an AI platform from Nvidia known as Isaac to train robots for certain new tasks.

“In the future, I am very positive that we can just put a new robot in this facility, and say, ‘OK, talk with the other robots and find the best way to produce this body,’” Grüeneisl says.

Manufacturers have used laptop simulations to hone their meeting strains for a while. But Omniverse permits the total manufacturing course of to be simulated with photo-realistic particulars, and with bodily properties like gravity and completely different supplies. It’s potential to lay out the manufacturing course of from begin to end, and to see how modifications to one half may need knock-on results on one other. It is less complicated to construct a extra advanced digital surroundings as a result of completely different 3D fashions will be imported into the system. Omniverse makes use of an open file normal appropriate with quite a few computer-aided design packages.

The software program can even simulate avatars of human employees grabbing elements and instruments, and assembling elements, to discover the finest process and reduce ergonomic issues. It may also make it potential for fewer employees to full a selected job, Grüeneisl says.

“We do AI simulation of how people move around the factory,” says Richard Kerris, common supervisor for Omniverse at Nvidia. He calls the challenge “one of the most complex simulations that’s ever been done.”

There’s rising curiosity in utilizing AI to management robots and different industrial machines. Encouraged by latest progress in AI, some startups are targeted on having robots be taught in simulation how to carry out fiendishly tough duties like grasping irregular objects, expertise that would ultimately assist automate a lot ecommerce and logistics work. This typically makes use of an AI strategy known as reinforcement learning, which entails an algorithm experimenting and studying, from constructive suggestions, how to obtain a particular objective.

“This is definitely the way to go,” says Ding Zhao, a professor at Carnegie Mellon University who focuses on AI and digital simulations. Zhao says simulations are essential to utilizing AI for industrial purposes, partly as a result of it’s unattainable to run machines by thousands and thousands of cycles to collect coaching knowledge. In addition, he says, it’s vital for machine-learning fashions to be taught by experimenting with unsafe conditions, corresponding to two robots colliding, which can’t be achieved with actual {hardware}. “Machine learning is data-hungry, and collecting it in the real world is expensive and risky,” he says.

Willy Shih, a professor at Harvard Business School who focuses on manufacturing expertise, says the sophistication of simulation has been growing steadily, and he says simulation primarily saves money and time by preempting future manufacturing issues.

Shih says there may be loads of hype round AI for manufacturing however provides, “There are lots and lots of applications” for the expertise too.

Nvidia CEO Jensen Huang mentioned BMW’s use of Omniverse throughout his keynote at the firm’s annual GTC convention, held nearly on Monday. Nvidia initially made graphics chips for gaming however broadened its focus when these chips proved adept at coaching AI applications. The firm has since leapt into a number of different industries the place AI is vital, together with automotive and medical imaging.