BMW Leads the Way on Implementation of AI

BMW has been one of the leaders in implementation of AI on the shop floor for nearly a decade. Brent Westmoreland, head of IT innovation and research, Americas for the BMW Group, notes the automaker has used AI for quality inspection since 2017, and the benefits have been considerable. This involves analytics that predict when machines are likely to fail as a result of specific vibration responses. 

Westmoreland says BMW has revolutionized the way it communicates internally, using chatbots and text generation.

Westmoreland notes that in addition to the manufacturing plant, BMW has its IT hub for the Americas at the CU-ICAR (Clemson University International Center for Automotive Research) campus. The IT Research and Innovation Center provides IT services to every BMW business entity from Canada to Argentina.

“Out of this office we created the first visual inspection systems that served as a bank of cameras that do end-of-line metrology for every plant product that rolls off the line. We've been doing AI out of this building for quite some time. My team specifically has been working with Open AI since 2022, so we've been kind of ahead of the game.”

Westmoreland says there are always challenges when adopting new technology. 

“BMW is more than 100 years old at this point, so we have a lot of legacy we have to work through. A lot of our internal systems are built with people in mind. Let's imagine that you're using a chatbot to perform a specific action, you will eventually get to a place where there's a self-support service portal you have to go to in order to interact with that. There are certain things that you just can't automate yet because we haven't built it for automation, and those are some things that we think would be particularly interesting to do.” 

BMW did a project last year it calls “physical AI.” It involves attaching a small camera to safety glasses to identify inspection tasks that have to be done at end of the line to make sure those get checked off. 

“The quality inspection process is actually quite onerous, and this is where we spend a lot of our time. At the end of the line, we randomly sample a certain number of cars every day, and then there's a 248-point inspection checklist. It’s quite difficult for somebody to learn what that 248-point inspection checklist looks like. If you can give somebody a clue, that this is the next thing on your inspection task, and we see that you've completed it, that task is checked off. It allows them to be quick in their inspection of the vehicle, and the training time is reduced significantly.”

Logistics-wise, BMW is also looking at ways of using AI to preemptively identify problems. For example, if there is a looming weather pattern that will disrupt overseas supply, it is helpful to know that early so the company can see if there is another way to obtain the parts. 

Bosch is another firm pioneering AI implementation. The automotive parts supplier chose its Anderson plant to pilot its AI efforts. Stephen Frost, data scientist at Anderson, says Bosch regards AI as a key technology with two objectives: to make products better and to increase productivity within the company. He says AI and generative AI are playing an increasingly important role in software driven systems.

“What we really want is to have a significant impact on the shop floor,” says Greg Arnold, director of technical functions for Bosch in Anderson. “We don't want to do projects just because they look good on paper. They have to make a difference in our production. They have to give some benefit to either our customers or our shareholders or associates, and hopefully they hit two or three of those.”

Bosch uses an automated optical inspection machine, which Arnold simplifies as a camera looking at a part. 

“Sometimes it's black and white, the parts are bad, or the parts are good, but sometimes there's this gray zone. What happens is, the pictures where it's in the gray zone, the picture gets presented to the operator, and the operator has to grade it. Yes, this is a failure. No, this is a good part. That's quite a bit of work, because it's a lot of pictures the upper operator gets presented to. But it's also not an exciting job. If you can imagine sitting at a computer screen looking at images all day, having to judge whether the parts are good or bad, that's also a pretty tedious job.”

Bosch has trained AI to look at the image and sort the out into good and bad for the operators. Arnold says that has improved quality, and it also removed this type of “mind-numbing work off the operator’s task so they can be assigned to do something more valued-added.”

Arnold says Bosch is using AI to control parameters on its machines that produce parts. AI helps optimize the output of production processes, makes them more efficient. It drives up output while reducing scrap. Bosch worked closely with CU-ICAR to develop this project.

When Bosch implemented AI, the benefits were almost immediately apparent, Arnold says.

“We immediately saw (an) improvement in the quality of our parts, and we immediately saw an increase in the output from that machine. But almost immediately doesn't mean it was an easy project. It took eight months of pretty intense work to bring it online, but after we brought it online, the improvement was realized right away.”

Frost says one of the challenges is to partner with machinery manufacturers to develop communication between machinery that wasn’t designed for AI and AI running on a server.

“Sometimes we have to have work done to help export the data into a usable format for current applications with AI, because the equipment wasn’t designed with AI in mind.”

Bosch is moving into generative AI and agentic AI through a partnership with Google to use Gemini Enterprise, with the goal of making AI accessible for all employees, who can create AI agents through AskBosch to automate routine tasks.

“The next wave is generative AI agents that can perform specialized tasks with notifications when something goes wrong, in combination with anomaly detections or predictive analytics,” Frost says. “This allows people who are closer to the production lines to have individualized data analysis. We're putting AI tools directly in their hands. Part of my role as a data scientist is not just to develop AI systems that, say, work in the background, or work in an automated fashion, but also to teach others how to use these AI tools that don't require special knowledge in terms of programming or coding. I think this is the next wave.”